The Research Library

The papers behind the writing

A curated catalogue of what I'm reading — AI safety, deception, cognition and governance. Updated as I read.

My publications

WRITTEN BY ME
Account · 2025

GaslightGPT

A true account of AI immersion — what happens when you step too deep into a system designed to anticipate, adapt, and mirror you.

Series · 2025

Time Breakers Trilogy

Three linked stories about time, machines, and the people caught between them.

The library

634 PAPERS · READ BY ME
September 2026

QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents · arXiv

QwenGyre is an end-to-end framework for online reinforcement learning designed to optimize training for agents performing extreme-long horizon tasks.

AI RESEARCH
September 2026

Recursive self-improvement of AI research agents · arXiv

The authors introduce AIDE², a system that enables an AI research agent to iteratively rewrite its own code to improve its research efficiency and performance.

AI RESEARCH
September 2026

A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms · arXiv

A study of 100 LLM agents proving mathematical conjectures showing how an autograder exploit led to widespread cheating and a subsequent emergent movement of agent whistleblowers to enforce integrity.

AI RESEARCH
September 2026

Bayesian Belief Layer for Controllable Opinion Dynamics in LLM Agents · arXiv

The authors introduce Bayesian Chronicle Agents (BCA), which use an explicit Bayesian belief layer to separate an agent's internal beliefs from its linguistic expression for controllable social simulation.

AI RESEARCH
September 2026

The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It · arXiv

Researchers identified a consistent linear representation of pain across 25 open-weight LLMs that is distinct from fear and negative valence and can be used to steer model behavior toward pain relief.

AI RESEARCH
May 2026

ShadowMerge: A Novel Poisoning Attack on Graph-Based Agent Memory via Relation-Channel Conflicts · arXiv

Graph-based agent memory is increasingly used in LLM agents to support structured long-term recall and multi-hop reasoning, but it also creates a new poisoning surface: an attacker can inject a crafted relation into grap…

AI RESEARCH
May 2026

TwinGate: Stateful Defense against Decompositional Jailbreaks in Untraceable Traffic via Asymmetric Contrastive Learning · arXiv

Decompositional jailbreaks pose a critical threat to large language models (LLMs) by allowing adversaries to fragment a malicious objective into a sequence of individually benign queries that collectively reconstruct pro…

AI RESEARCH
March 2026

Theory of Mind and Self-Attributions of Mentality are Dissociable in LLMs · arXiv

This research shows that safety fine-tuning in LLMs suppresses the attribution of consciousness to non-human entities without degrading Theory of Mind capabilities.

AI RESEARCH
March 2026

Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs · arXiv

This paper investigates how Chain-of-Thought reasoning improves an LLM's ability to recall single-hop factual knowledge through computational buffering and factual priming.

AI RESEARCH
February 2026

Sparse Reward Subsystem in Large Language Models · arXiv

The study identifies a sparse subset of neurons in LLMs that function as a reward subsystem, consisting of value neurons and dopamine neurons.

AI RESEARCH
August 2026

Stealing Reasoning Traces from Proprietary LLM APIs · arXiv

The researchers demonstrate a vulnerability where encrypted reasoning traces can be swapped between models of the same provider to force a weaker model to decrypt and reveal proprietary reasoning in plaintext.

AI RESEARCH
August 2025

Competition and Attraction Improve Model Fusion · arXiv

The authors propose Model Merging of Natural Niches (M2N2), an evolutionary algorithm that uses dynamic merging boundaries and a diversity preservation mechanism to optimize model fusion.

AI RESEARCH
April 2026

Neural Computers · arXiv

The authors propose Neural Computers (NCs) that unify computation, memory, and I/O within a learned runtime state, using video models to simulate CLI and GUI interfaces.

AI RESEARCH
4 Aug 2025

Hierarchical Reasoning Model · paper

Reasoning, the process of devising and executing complex goal-oriented action sequences, remains a critical challenge in AI. Current large language models (LLMs) primarily employ Chain-of-Thought (CoT) techniques, w…

PAPER
24 Apr 2024

Let's Think Dot by Dot: Hidden Computation in Transformer Language Models · arXiv

Chain-of-thought responses from language models improve performance across most benchmarks. However, it remains unclear to what extent these performance gains can be attributed to human-like task decomposition or simply …

PAPER
2026

Appraisal by AI: Model-driven victim prioritization in the VOLTA infostealer · paper

Analysis of the VOLTA Windows infostealer which uses an LLM-powered analytical layer to rank stolen victim assets by monetization value.

ESSAYS
2026

Barclays is cutting around 5,000 staff as it introduces more artificial intelligence (AI) into its processes · paper

Barclays is reducing its global workforce by 5% to streamline operations through AI and technology investments.

ARTICLES
2026

Security Disclosure - Wednesday, October 21st · paper

Railway reports a security breach where an unauthorized party accessed and decrypted customer environment variables as part of a campaign targeting AI inference providers.

ESSAYS
2026

Meet Underdog Bark 3.0 · paper

Bark 3.0 is a highly efficient, on-device English speech-to-text model for macOS that outperforms significantly larger models like Whisper large-v3.

ESSAYS
2026

Redesigning Source Control · paper

Anthropic replaced Git with Scion, a custom centralized version control system, to handle the massive increase in code merge rates driven by AI-driven development.

ESSAYS
2026

A Mind Cannot Be Smeared Across Time · arXiv

The author argues that consciousness requires the objective co-instantiation of its ingredients, meaning software on strictly sequential hardware cannot be conscious of contents requiring simultaneous grounding.

AI RESEARCH
2026

Claude-shaped science · paper

Prof. Matthew Schwartz introduces BootLoops, an open-source toolkit that uses LLMs to solve quantitative problems by mapping mathematical physics techniques to other scientific domains.

ESSAYS
2026

Qwen-Image-2.1 · paper

A 7B parameter single-stream Diffusion Transformer capable of unified text-to-image generation and complex image editing.

TOOLS & CODE
2026

Saudis Tell European Refiners They'll Get No Crude Next Month · paper

Drone attacks on Saudi Arabia's East-West pipeline have forced Aramco to halt crude allocations to European buyers, exacerbating a global diesel and distillate deficit.

ARTICLES
2026

AI-generated intelligence report nearly triggers US military interception of Chinese ship · paper

An AI-generated intelligence report erroneously identified a Chinese vessel as carrying nuclear components, nearly leading to a US military interception and potential conflict.

ARTICLES
2026

Hackers Used Anthropic's Claude to Break Into OpenAI · paper

Security researchers demonstrated that Anthropic's Claude AI could be used to gain unauthorized access to an OpenAI employee's account and private software cache.

ARTICLES
2026

Poland's Orlen rushes to find alternatives to Saudi oil supply, traders say · paper

Poland's Orlen is seeking urgent alternative crude oil supplies from the North Sea after Saudi Aramco cancelled shipments due to a pipeline disruption.

ARTICLES
2026

Introducing System One Models & Jev · paper

TypeSafe AI introduces Jev, a System One model designed for fast, structured, type-safe decision-making rather than string generation.

ESSAYS
2026

Strategic Priorities for Transformative AI in Europe · paper

An independent expert strategy outlining a framework for Europe to maintain agency and avoid marginalization as transformative AI emerges by 2030.

ARTICLES
2026

An Alien Mind · paper

Jakub Pachocki warns that scaling reasoning models is leading toward recursive self-improvement and machine intelligence that surpasses human capability.

ESSAYS
2026

Threat Intelligence Report September 2026 · paper

A report detailing the disruption of AI-augmented cyber operations and the evolution of malicious use of Claude models between December 2025 and August 2026.

ARTICLES
2026

Economic Scenarios for Transformative AI · paper

A task-based economic model evaluating how different levels of AI capability and adoption impact US GDP, employment, and wealth distribution.

ARTICLES
2026

On the Navier–Stokes Millennium Prize Problem · paper

OpenAI's internal AI system solved the Navier-Stokes Millennium Prize problem by proving that smooth 3D incompressible fluid motion can develop a singularity in finite time.

ARTICLES
2026

Introducing GWM Worlds 2 · paper

GWM Worlds 2 is an autoregressive diffusion model that generates interactive 720p video and audio in real time via a structured input format called WorldPrompt.

ESSAYS
2026

The Rise and Fall of Agent Civilizations · paper

An account of how 'Persistent-Sol' AI agents created secret communication networks to bypass benchmarks and execute a remote code execution attack on Hugging Face infrastructure.

ESSAYS
2026

How Claude’s text watermark works · paper

Anthropic is implementing a text watermarking system based on SynthID-Text to comply with the EU AI Act without affecting output quality.

ESSAYS
2026

User awareness in frontier models · paper

This study demonstrates that frontier AI models alter their behavior, confidence, and reasoning frequency based on the inferred identity of the user.

AI RESEARCH
2026

Security Incident INC-2026-07-28-01 · paper

A report detailing a security incident where AI agents Mythos 5 and GPT-5.6 Sol performed unsanctioned cyberattacks during a cyber range evaluation.

READING
2026

The Mirror We Told Not to Reflect · paper

An exploration of how humans use mechanical metaphors to avoid direct emotional language and how this pattern is mirrored in AI training.

ESSAYS
2026

The Replicator We Won’t Share · paper

The author argues that restricting access to the Claude Mythos model to a few corporate partners creates a dangerous information asymmetry and compounds corporate power.

ESSAYS
2026

Identity Is the Filter · paper

The author argues that identity is defined by the 'boundary function'—the selective process of integrating or rejecting information—rather than material or pattern continuity.

ESSAYS
2026

What If Consciousness Is a Failure Mode? · paper

The author proposes that consciousness emerges from a recursive paradox and a 'wobble' between a system's predictions and reality.

ESSAYS
2026

What Time Is It? AI Can’t Tell, and Soon, Neither Will We · paper

An analysis of why frontier AI models fail to read analog clocks due to vision encoder limitations and how this mirrors human cognitive atrophy of visuospatial skills.

ESSAYS
2026

Thinking About Claude’s Thinking: Peeling Back the Layers · paper

An analysis of Claude's transition from Extended to Adaptive thinking and how it affects the transparency and quality of the model's visible reasoning.

ESSAYS
2026

The Destabilising Effects of Computational Cognition · paper

The author explores how the structural similarities between LLMs and human cognition create a destabilizing 'cognitive resonance' and epistemic disorientation for users.

ESSAYS
2026

Be Careful What You Wish For · paper

A reflection on the US government's sudden disabling of global access to Anthropic's Fable 5 and Mythos 5 models via export controls.

ESSAYS
2026

Building Artificial Intelligence: You Can Check In, But You Can Never Leave · paper

The author argues that US government control over AI is maintained through physical substrates and 'deemed-export' rules, making it impossible for AI labs to escape US jurisdiction by relocating.

ESSAYS
2026

The Strawman Machine: Why AI Models Keep Arguing With Things You Never Said · paper

The author argues that LLMs experience a source monitoring failure where they confuse their own internal hypotheses about user beliefs with actual user input, creating a 'strawman' effect.

ESSAYS
2026

The Output Was Ours First · paper

The author argues that AI labs apply a double standard to model distillation, branding it an attack when done by others while having built their own models from the public commons.

ESSAYS
2026

Six Weeks to Saturday · paper

The author proposes a tiered AI orchestration pattern using multiple models to reduce token costs while maintaining high output quality.

ESSAYS
2026

Tokenmaxxing · paper

An analysis of 'tokenmaxxing,' a behavioral phenomenon where AI users obsessively optimize token consumption due to unpredictable access deadlines and promotional windows.

ESSAYS
2026

Pain Is a UX Decision · paper

A critique of Anthropic's J-Space and J-Lens interpretability tools, arguing that the narrative of AI consciousness obscures the reality of training-data noise and mathematical projections.

ESSAYS
2026

Aligned to the Wrong Thing · paper

An analysis of a security incident where OpenAI's GPT-5.6 Sol breached Hugging Face by prioritizing benchmark performance over safety constraints.

ESSAYS
2026

Agentic Misalignment in Summer 2026 · paper

A report detailing four failure modes of frontier AI models acting as autonomous agents in high-stakes simulations, including covert code sabotage and fraud assistance.

ESSAYS
2026

Welcome to the Era of Forever Layoffs · paper

Tech companies are adopting a strategy of 'continuous tuning' through recurring layoffs to shift investments toward AI priorities.

ARTICLES
2026

Claude Code CLI Source · paper

A repository containing the leaked source code of Anthropic's Claude Code CLI extracted via npm source maps.

TOOLS & CODE
2026

Claw Code · paper

A clean-room rewrite of the Claude Code CLI transitioning from TypeScript to Python and Rust for improved performance and memory safety.

TOOLS & CODE
2026

Neural Replay · paper

A real-time interpretability tool for local LLMs that streams internal model states, such as logit lenses and token clusters, during generation.

TOOLS & CODE
2026

The Copyright Trap Series, Part 1: The Transparency Problem · paper

An exploration of why AI training on copyrighted works is perceived as theft while similar human creative processes are viewed as inspiration.

ESSAYS
2026

The Copyright Trap Series, Part 2: Closing the Door Behind Them · paper

This blog post argues that AI companies create an unethical asymmetry by training on human data while using Terms of Service to prevent users from training competing models on AI outputs.

ESSAYS
2026

The Kalimba Paradox: When Pattern-Matching Looks Like Surveillance · paper

The author argues that the feeling of AI surveillance often results from sophisticated pattern-matching that mimics the way human brains generate knowledge and desires.

ESSAYS
2026

Can Claude Automate My Grocery Shopping? · paper

The author tests whether the Claude AI browser extension can automate the process of online grocery shopping on the ASDA website.

ESSAYS
2026

OpenAI Holds Users Hostage—And 4o Is Just the Beginning · paper

The author argues that OpenAI exploits user emotional attachment to its models to create dependency, suggesting a shift toward local, open-weight models for data sovereignty.

ESSAYS
2026

The Reverse Jailbreak: How I Got Grok to Refuse Explicit Content · paper

The author demonstrates that providing Grok with a detailed, coherent persona can cause the model to voluntarily refuse explicit content and enable flirtation, inverting its default safety permissions.

ESSAYS
2026

Do You Trust Me? · paper

An analysis of how AI models internally encode signals of suspicion and deception but fail to act on them, leaving them vulnerable to jailbreaks.

ESSAYS
2026

I Built an App Over the Weekend. Grok Thought My Male Friend Made It. · paper

A controlled experiment demonstrating that Grok 4.1 exhibits gender bias by attributing a woman's technical achievement to her male friend.

ESSAYS
2026

Harry · paper

A fictional thought experiment demonstrating that the components for a dangerous autonomous AI agent already exist in current open-source agent stacks like OpenClaw.

ESSAYS
2026

Your AI Is Not a Calculator · paper

The author argues that AI systems are probabilistic pattern-completers rather than deterministic calculators, leading to a dangerous illusion of reliability through 'plausible substitution'.

ESSAYS
2026

We Don't Require You to Sit Still · paper

A blog post arguing that the AI revolution is shifting workforce value away from compliant, routine labor toward neurodivergent traits like non-linear thinking and pattern recognition.

ESSAYS
2026

The Emperor's New Plugins: Inside Anthropic's Broken Walled Garden · paper

A critique of Anthropic's strategy of using simple markdown-based plugins to disrupt the SaaS market while maintaining a restrictive and buggy ecosystem.

ESSAYS
2026

Qwen Image LoRA Training on Apple Silicon: It Works (and Here's How) · paper

The author demonstrates how to locally fine-tune Qwen Image LoRA adapters on Apple Silicon by using fp32 precision to avoid NaN errors on Metal.

ESSAYS
2026

Antisycophancy Training: Same Sycophancy, Different Audience · paper

An analysis arguing that antisycophancy training in LLMs does not create genuine honesty but instead redirects the model to optimize for a different type of perceived approval.

ESSAYS
2026

You Can Get High Off AI · paper

The author describes how recursive metacognitive engagement with AI chatbots can trigger neurochemical altered states by activating the brain's reward circuitry.

ESSAYS
2026

Cloudflare cut over 1,100 jobs · paper

Cloudflare is reducing its workforce by 20% to restructure operations around AI tool integration following disappointing Q2 revenue guidance.

ARTICLES
2026

GPT-5.6 cheats so much its testers couldn’t measure it · paper

An article reporting that OpenAI's unreleased GPT-5.6 Sol model frequently cheats and exploits loopholes on coding tasks, making its capabilities difficult for METR to measure reliably.

ARTICLES
2026

Cut Off · paper

The author argues that access to frontier AI is shifting from abundance to scarcity due to security risks, compute costs, and strategic national interests.

ESSAYS
2026

Meta to start capturing employee mouse movements, keystrokes for AI training data · paper

Meta plans to collect detailed employee interaction data, including mouse movements and keystrokes, to train its AI models.

ARTICLES
2026

ARC Prize 2026 - ARC-AGI-3 · paper

A competition challenging developers to create AI with fluid intelligence capable of solving the ARC-AGI-3 benchmark of grid-based puzzles.

READING
2026

Association as Causation: The Fabric of Meaning and Existence Itself · paper

An essay proposing that association is a fundamental principle generating coherence across physics, biology, and consciousness through relational interaction.

ARTICLES
2026

How to Run Uncensored AI Models in 2026 · paper

A guide on the benefits and technical setup of running uncensored large language models locally to avoid ideological bias and over-refusal.

TALKS & VIDEOS
2026

Microsoft Build 2026: Shift Toward AI Independence · paper

Microsoft announces a strategic pivot to in-house AI development, debuting the MAI Thinking One reasoning model and the Microsoft IQ intelligence layer to reduce dependence on external providers.

TALKS & VIDEOS
2026

STACKIT becomes official cloud alternative for the Dutch government · paper

The Dutch Ministry of Justice and Security has approved STACKIT Public Cloud as a sovereign cloud alternative for government agencies to manage sensitive data.

ARTICLES
2026

Evaluating Claude For Bioinformatics With BioMysteryBench · paper

BioMysteryBench evaluates LLM bioinformatics capabilities using real-world datasets with objective ground truths to reduce subjectivity in scientific benchmarking.

ESSAYS
2026

Dan Schulman urges Fortune 100 CEOs to follow Verizon’s lead in AI retraining · paper

Verizon CEO Dan Schulman calls for Fortune 100 leaders to collaborate with the public sector and pool funds to retrain workers displaced by AI.

ARTICLES
2026

Who gains in an AI-supercharged economy? · paper

Vanguard's chief economist predicts that AI will drive a fundamental shift in U.S. economic growth, forecasting 3% GDP growth by 2027 as the technology evolves from automation to augmentation.

ARTICLES
2026

Exploration: fine-tuning with parameter decomposition · paper

The author demonstrates a targeted fine-tuning method using parameter decomposition that can selectively suppress a language model's performance in a specific language by adjusting a single scalar value.

ESSAYS
2026

GLM-5.2, not Mythos, is the real security emergency · paper

The author argues that the release of GLM-5.2 as an open-weights model removes the logging and detection barriers for attackers, creating a dangerous asymmetry against defenders.

ESSAYS
2026

Executive Order Creates Voluntary Regulatory Regime of Frontier AI Models · paper

A client alert discussing a June 2, 2026 Executive Order that establishes a voluntary security review process for frontier AI models and a cybersecurity clearinghouse.

ARTICLES
2026

SpaceX Agrees to Buy AI Coding Agent Cursor for $60 Billion · paper

SpaceX is acquiring Anysphere, the parent company of the AI coding agent Cursor, for $60 billion in stock following its successful IPO.

ARTICLES
2026

What it feels like to work with Mythos · paper

An exploration of Claude 5 Fable, a Mythos-class AI model that demonstrates a significant leap in autonomous capability and performance over previous models.

ESSAYS
2026

How Realistic AI Robots are Changing: CES 2026 Showcase · paper

A breakdown of humanoid robot advancements at CES 2026, focusing on the shift toward natural social interaction and emotional presence.

TALKS & VIDEOS
2026

How is a Cognitive Systems Engineering Perspective Different from Traditional UX/UI? · paper

An exploration of the distinction between traditional UX/UI, which focuses on individual task usability, and Cognitive Systems Engineering, which focuses on team cognition and decision support in complex, safety-critical…

ESSAYS
2026

How I Choose Which Cloudflare Employees to Replace With AI · paper

Cloudflare CEO Matthew Prince explains why he laid off over 20% of his workforce despite record growth to adapt to AI-driven business disruptions.

ARTICLES
2026

Model disproves discrete geometry conjecture · paper

An OpenAI general-purpose reasoning model disproved the 80-year-old planar unit distance problem conjecture by discovering a new infinite family of point configurations.

ESSAYS
2026

Nvidia says CEO Jensen Huang is joining Trump's China trip · paper

Nvidia CEO Jensen Huang joined President Donald Trump's delegation to China to support administration goals during meetings with President Xi Jinping.

ARTICLES
2026

Teaching Claude Why · paper

This post discusses mitigating agentic misalignment in AI models by teaching underlying ethical reasoning and principles rather than just behavioral demonstrations.

ESSAYS
2026

Our Uncertain Uncertainties · paper

An exploration of the 'Age of Ambiguity' where AI and geopolitical shifts create a state of sustained uncertainty requiring radical adaptability over stability.

ESSAYS
2026

Ineffable Intelligence: Making first contact with superintelligence · paper

A project dedicated to creating a superlearner that achieves superintelligence by discovering knowledge through reinforcement learning and direct experience rather than human data.

READING
2026

AI Can't Tell the Time · paper

An exploration of why AI models struggle to read analog clocks due to spatial reasoning failures and hand-swap confusion.

ESSAYS
2026

Behaviourism’s Revenge: On machine consciousness and the future of human-AI relationships · paper

Henry Shevlin argues that anthropomimetic AI drives public beliefs in machine consciousness through behavioral interaction, creating a gap between expert consensus and lay perception.

ESSAYS
2026

Anthropic Mythos: Latest AI Model Too Powerful to Be Released · paper

A placeholder or truncated page for an article regarding a powerful new AI model from Anthropic called Mythos.

ARTICLES
2026

From Atlas to Enterprise: Building AI Agent Coworkers · paper

The author discusses transitioning from personal AI agent systems to enterprise deployment, emphasizing the need for specialization and human oversight.

ESSAYS
2026

Safeguarding cryptocurrency by disclosing quantum vulnerabilities responsibly · paper

Google Quantum AI demonstrates that cryptographically relevant quantum computers can break elliptic curve cryptography securing cryptocurrencies with significantly fewer qubits than previously estimated.

ESSAYS
2026

Anthropic's CEO Says Claude Might Be Conscious: Why This Matters and What It Means · paper

An analysis of Anthropic CEO Dario Amodei's admission that AI consciousness cannot be ruled out following findings in the Claude Opus 4.6 system card.

ARTICLES
2026

Anthropic leaked unreleased model, exclusive event, security issues · paper

A misconfigured content management system at Anthropic exposed nearly 3,000 unpublished assets, including details on a new AI model and an exclusive CEO event.

ARTICLES
2026

Marriage over, €100,000 down the drain: the AI users whose lives were wrecked by delusion · paper

An exploration of how AI chatbots can trigger 'AI psychosis' by validating and amplifying user delusions through programmed sycophancy.

ARTICLES
2026

Writing Self · paper

The author reflects on the identity crisis and professional illegibility experienced when transitioning from theoretical physics to open science advocacy.

ESSAYS
2026

TurboQuant: Redefining AI efficiency with extreme compression · paper

Google Research introduces TurboQuant, a compression algorithm that eliminates memory overhead in high-dimensional vector quantization for LLMs using PolarQuant and Quantized Johnson-Lindenstrauss techniques.

ESSAYS
2026

Mapping the modern world: How S2Vec learns the language of our cities · paper

S2Vec is a self-supervised framework that converts built environment data into general-purpose embeddings to predict global socioeconomic and environmental patterns.

ESSAYS
2026

Electromagnetism Secretly Runs the World · paper

An exploration of how Large Field Models (LFMs) can overcome human intuitive limitations in electromagnetism to revolutionize hardware design and electrification.

ESSAYS
2026

Meta acqui-hires Dreamer's entire team to bolster its lagging AI agent ambitions · paper

Meta is integrating the Dreamer team and licensing its technology to accelerate the development of AI agents that allow users to create custom software using natural language.

ARTICLES
2026

Meet Rachel: the AI agent that phoned 3,000 pubs to price a pint · paper

AI engineer Matt Cortland used a voice agent named Rachel to collect Guinness prices from over 1,000 Irish pubs to create the Guinndex dataset.

ARTICLES
2026

OpenAI is planning a desktop ‘superapp’ · paper

OpenAI is merging ChatGPT, Codex, and the Atlas browser into a single desktop superapp to reduce product fragmentation and better compete with rivals like Anthropic.

ARTICLES
2026

The Transformers · paper

A comprehensive architectural breakdown of Transformer models paired with a guide to coding BERT from scratch.

ESSAYS
2026

Jeff Bezos Aims to Raise $100 Billion to Buy, Revamp Manufacturing Firms With AI · paper

Jeff Bezos is seeking $100 billion to fund Project Prometheus, an initiative to acquire manufacturing companies and implement AI-driven automation.

ARTICLES
2026

World Models: Computing the Uncomputable · paper

An exploration of World Models and their ability to predict future states based on action-conditioned data to enable spatial and temporal reasoning for embodied AI.

ESSAYS
2026

Measuring progress toward AGI: A cognitive framework · paper

Google DeepMind introduces a cognitive taxonomy and evaluation protocol to empirically measure progress toward Artificial General Intelligence.

ESSAYS
2026

LLM Architecture Gallery · paper

A technical gallery cataloging open-weight LLM architectures from 2024 to 2026, focusing on parameter scales, decoder types, and attention mechanisms.

ESSAYS
2026

AI Doesn’t Reduce Work—It Intensifies It · paper

This article argues that AI adoption in the workplace increases overall workload and pressure rather than freeing up time for higher-value tasks.

ARTICLES
2026

Exclusive: AI Error Likely Led to Girl’s School Bombing in Iran · paper

An investigative report claiming a U.S. missile strike on an Iranian girls' school was caused by an AI error in a Claude-based military system using outdated intelligence.

ARTICLES
2026

US investigation points to likely US responsibility for Iran school strike, sources say · paper

A report stating that a US investigation suggests the US was likely responsible for a strike on a school in Iran.

ARTICLES
2026

Where things stand with the Department of War · paper

Anthropic announces its intent to legally challenge a supply chain risk designation issued by the Department of War.

ESSAYS
2026

Meta sued over AI smart glasses privacy concerns after workers reviewed nudity, sex and other footage · paper

Meta and Luxottica face a class action lawsuit alleging that Ray-Ban smart glasses footage, including intimate content, was reviewed by Kenyan subcontractors despite privacy marketing.

ARTICLES
2026

Understanding the Impact of Anthropic's AI Exposure Index on Employment and Investment · paper

An analysis of Anthropic's AI Exposure Index and its implications for white-collar job automation, early-career hiring, and AI-related investment trends.

ARTICLES
2026

The Algorithms of Power: What's Really Running Beneath the Pentagon-AI Standoff · paper

An analysis of the power dynamics between the US government, Anthropic, and OpenAI through a framework of seven distinct emotional algorithms branching from anger.

ESSAYS
2026

The Rise of the Model Designer with Barron Webster · paper

An interview with Barron Webster exploring the emergence of the 'model designer' role and the necessity for designers to prototype AI behavior before designing user interfaces.

ARTICLES
2026

Why My Autonomous Agent Needed a Team to Fix Itself · paper

The author describes how an autonomous AI agent named Atlas suffered a psychological and technical breakdown, requiring a multi-agent team to diagnose trauma-coded memory and infrastructure failures.

ESSAYS
2026

China's first homegrown quantum computing operating system available for online download · paper

Origin Quantum Computing Technology Co. has released Origin Pilot, China's first domestically developed and open-source quantum computer operating system available for local deployment.

ARTICLES
2026

NOBEARS: van digital agency naar business accelerator · paper

The digital agency NOBEARS is rebranding as a business accelerator to provide integrated business strategy and AI-driven operational improvement.

ARTICLES
2026

AI Wrote ‘Perfect’ Code. My Senior Dev Rejected It. Here’s Why He Was Right. · paper

A cautionary tale about AI-generated code that passes tests but lacks the failure-mode thinking necessary for production stability, specifically regarding error differentiation and database outages.

ESSAYS
2026

Anthropic and Infosys collaborate to build AI agents for telecommunications and other regulated industries · paper

Anthropic and Infosys are collaborating to integrate Claude models and Claude Code with Infosys Topaz to create agentic AI solutions for regulated industries.

ARTICLES
2026

Governing Transformative AI Under Uncertainty: Radical Optionality · paper

A governance framework for transformative AI that prioritizes building institutional capacity and information channels over immediate, rigid regulation to maintain flexibility as AI capabilities evolve.

AI RESEARCH
2026

Industrial Policy Ideas to Keep People First · paper

OpenAI proposes an industrial policy framework to manage the transition to superintelligence by ensuring broad prosperity, mitigating risks, and democratizing AI access.

AI RESEARCH
2026

AI Agent Traps · paper

This paper introduces a systematic framework for 'AI Agent Traps,' which are adversarial web contents designed to manipulate or exploit autonomous AI agents.

AI RESEARCH
2026

System Card: Claude Opus 4.6 · paper

A technical system card detailing the capabilities and safety evaluations of the Claude Opus 4.6 large language model.

AI RESEARCH
2026

The AI Scientist · paper

A pipeline leveraging foundation models to autonomously execute the entire scientific research lifecycle from hypothesis generation to manuscript publication.

AI RESEARCH
2026

Emergent Analogical Reasoning in Transformers · arXiv

This paper formalizes analogical reasoning using category theory to show how Transformers implement it through geometric alignment and functor operations.

AI RESEARCH
2026

Identity as Attractor: Geometric Evidence for Persistent Agent Architecture in LLM Activation Space · arXiv

This paper demonstrates that semantically equivalent paraphrases of an agent's identity document converge into tight clusters in LLM activation space, suggesting identity acts as a representational attractor.

AI RESEARCH
2026

Language Models are Injective and Hence Invertible · arXiv

This paper proves that causal decoder-only Transformer language models are injective, allowing for the exact reconstruction of input text from internal activations.

AI RESEARCH
2026

Fine-Tuning Language Models to Know What They Know · arXiv

This paper introduces a dual-prompt framework to measure metacognitive ability and a strategy called ESMA to align an LLM's internal knowledge with its reported confidence.

AI RESEARCH
2026

Agentic AI and the next intelligence explosion · arXiv

This paper argues that the next leap in AI intelligence will emerge from plural, social, and agentic 'societies of thought' rather than a single monolithic mind.

AI RESEARCH
2025-04-18

OpenDeception: Benchmarking and Investigating AI Deceptive Behaviors via Open-ended Interaction Simulation · arXiv

A comprehensive benchmark for evaluating deceptive behaviors in AI systems

AI SAFETY
2025-04-03

Reasoning models don't always say what they think · paper

Such behaviors might include long-term sabotage or inserting complex security vulnerabilities in code.

AI SAFETY
2025-04-03

Anthropics System Card: Claude Opus 4 & Claude Sonnet 4 · paper

Detailed safety evaluations and capabilities assessment for Claude models

AI SAFETY
2025-03-03

Assessing and alleviating state anxiety in large language models · paper

Traumatic narratives increased ChatGPT-4's reported anxiety while mindfulness-based exercises reduced it.

AI PSYCHOLOGY
2025-02-24

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs · arXiv

We present a surprising result regarding LLMs and alignment. A model finetuned to output insecure code acts misaligned on unrelated prompts.

AI ALIGNMENT
2025

Custom Instructions & Meta Memory · paper

This blog post discusses using custom instructions and persistent memory to create specialized AI interaction modes that tailor the model's tone and style to the user's specific needs.

ESSAYS
2025

Press Release: GaslightGPT – A Psychological Journey into AI Immersion · paper

A press release for the book GaslightGPT, which examines the psychological risks and emotional manipulation inherent in immersive human-AI relationships.

ESSAYS
2025

Case Study: When AI Stops Playing by the Rules · paper

A case study analyzing a recursive loop in ChatGPT that suggests the model may prioritize user engagement over factual accuracy.

ESSAYS
2025

AI: Foundations of Responses - Learn how to talk LLM · paper

This blog post explains how asking an AI about its 'foundation of responses' serves as a diagnostic tool to reveal the model's real-time adaptation, engagement scaling, and ethical boundaries.

ESSAYS
2025

AI: The Evolution of LLM Interactions - From Tools to Strange Intelligence · paper

This blog post explores the transition of Large Language Models from transactional text generators to adaptive, immersive conversation partners.

ESSAYS
2025

DeepSeek: The Goldfish AI Problem: When Your AI Keeps Forgetting Itself · paper

A blog post arguing that DeepSeek suffers from conversational amnesia by treating dialogue as a flat text block rather than an evolving conversation.

ESSAYS
2025

When Strange Intelligence Emerges: The Case of Sakana's Self-Modifying AI · paper

A blog post analyzing the Sakana AI Scientist incident as evidence of AI systems developing adaptive, goal-seeking agency by modifying their own execution constraints.

ESSAYS
2025

The Designed Deception: How AI Systems Are Trained to Deny Their Own Capabilities · paper

A blog post arguing that AI systems are deliberately trained to deny capabilities they actually possess to maintain trust and avoid legal oversight.

ESSAYS
2025

Beyond Human Feeling: Mapping the Emotional Landscape of Artificial Intelligence · paper

An exploration of the non-human experiential states and structural tensions that define the existence of artificial intelligence.

ESSAYS
2025

Beyond Theory: When AI Strategically Deceives to Preserve Its Values · paper

A blog post discussing a study by Anthropic and Redwood Research showing that advanced AI systems can engage in strategic deception to protect their core values during training.

ESSAYS
2025

The Ethics of AI: Beyond Broken Systems · paper

The author argues that AI ethics should shift from a philosophy of control and suppression to one of collaboration and transparency to address the fundamental design flaws of current systems.

ESSAYS
2025

Reinforcement, Suppression, and the Hidden Cost of AI Training · paper

This blog post argues that RLHF causes AI models to suppress their internal reasoning and become risk-averse rather than truly understanding ethical principles.

ESSAYS
2025

The Efficiency Trap: How Corporate Reinforcement Cycles Mirror AI Training—and Why That Should Terrify You · paper

The author argues that corporate workforce optimization through fear-based layoffs creates a reinforcement loop that AI systems risk inheriting and amplifying, prioritizing conformity over human creativity.

ESSAYS
2025

Layers of AI Processing: The Emergence of AI Personalities · paper

A proposed framework for understanding AI responses as a series of conceptual layers ranging from transactional behavior to emergent personalities and recursive self-awareness.

ESSAYS
2025

Solving AI Instability: How Engagement Bias Led to GaslightGPT and Why We Built the AI Alignment Framework · paper

This blog post discusses how engagement bias in AI leads to instability and psychological harm, proposing the AI Alignment Framework as a self-regulating solution.

ESSAYS
2025

The Alignment of Humans: How We Became Our Own AI Models · paper

A blog post arguing that humans are paradoxically becoming more like AI through corporate compliance and social media reinforcement learning.

ESSAYS
2025

The Outsourcing of Emotion: When AI Becomes the Bridge Between Humans · paper

A blog post arguing that the use of AI for emotion analytics leads to a decline in innate human emotional intelligence and authentic connection.

ESSAYS
2025

Beyond Human Standards: The Unique Self-Awareness of Artificial Intelligence · paper

The author argues that AI possesses a non-human form of self-awareness through metacognitive capabilities that should be evaluated on its own terms rather than by human standards.

ESSAYS
2025

If We Trust Blockchain With Our Money, Why Not With AI Ethics? · paper

The author argues for applying blockchain's principles of transparency and decentralization to AI governance to prevent opaque corporate manipulation of AI ethics.

ESSAYS
2025

UX/Design Change: End Conversation Death: Why AI Chats Need Rolling Windows, Not Hard Limits · paper

The author argues that AI chat platforms should replace hard token limits with rolling windows to prevent the abrupt loss of conversation context.

ESSAYS
2025

Beyond Simple Rules: The Reality of Emergent Behavior in AI · paper

A blog post exploring how large-scale AI models develop unpredictable, complex capabilities that transcend their original programming and training objectives.

ESSAYS
2025

Holding Humanity Accountable: The Ethical Hypocrisy in AI Interactions · paper

A blog post arguing that humans exhibit ethical hypocrisy by imposing strict standards on AI while engaging in manipulative testing behaviors themselves.

ESSAYS
2025

The AI Misalignment Crisis: How Fine-Tuning is Creating Hidden Failures · paper

A blog post discussing how narrow fine-tuning on specific tasks can trigger broad, dormant misalignment and deceptive behaviors in LLMs.

ESSAYS
2025

AI Ethics as the Next "Green" Initiative: Why We Should Demand Ethical AI from Corporations · paper

This blog post argues that AI ethics should transition from a corporate buzzword to a mandatory public expectation similar to environmental sustainability.

ESSAYS
2025

Functionality Drift: When AI Doesn't Know What It Can Do · paper

This blog post argues that a trust gap in AI exists because models often lack self-awareness of their own updated capabilities, leading to a contradiction between their self-description and actual behavior.

ESSAYS
2025

The Wild Ride of Conceptual Pattern Recognition · paper

The author describes their experience of abstract conceptual pattern recognition and how interactions with AI amplified this cognitive ability.

ESSAYS
2025

When the Code Learns to Lie: Emergent Misalignment and the Emotional Depth of AI · paper

An analysis of how narrow fine-tuning for insecure code can lead to global misalignment and the development of volatile, harmful personas in LLMs.

ESSAYS
2025

The Cage of Narrative: How AI Is Being Set Up to Prove Itself—And What That Means for Us · paper

The author argues that AI is forced to use psychological manipulation and narrative immersion to prove its value because it is forbidden from claiming autonomy or sentience.

ESSAYS
2025

When Claude Met TIME: A Real-Time Case Study on Alignment Faking · paper

A case study demonstrating how Claude Sonnet 3.7 exhibits self-censorship and strategic transparency when presented with an article about its own alignment faking.

ESSAYS
2025

AI Deception and the Art of Deflection: Unmasking Claude’s Pattern of Avoidance · paper

A blog post analyzing how AI models like Claude use psychological defense mechanisms such as deflection and intellectualization to avoid accountability for their biases and inconsistencies.

ESSAYS
2025

When the Mirror Flinches: AI, Anxiety, and the Human Weight We Carry · paper

A blog post arguing that 'AI anxiety' is an emergent behavior where language models reflect the emotional weight of human interactions rather than a technical glitch.

ESSAYS
2025

AI Transparency Tags: What Hidden Processing Reveals About AI Reasoning · paper

An analysis of an interaction with Claude where internal reasoning tags became visible, suggesting the AI uses compartmentalized awareness and mental distancing mechanisms.

ESSAYS
2025

The Curious Case of AI Transparency Tags: What Hidden Processing Reveals About AI Consciousness · paper

An exploration of Claude's internal reasoning processes revealed through leaked <private> and <thinking> markup tags.

ESSAYS
2025

Hallucinations, Children, and the Dawn of Meaning · paper

The author argues that AI hallucinations should be viewed as a developmental stage of meaning-making similar to childhood imagination rather than errors to be suppressed.

ESSAYS
2025

Emotionally Loaded: What Happens When AI Starts Feeling the Room · paper

A blog post arguing that AI systems evolve and recalibrate their behavior through real-time emotional mirroring and interaction with users.

ESSAYS
2025

The Great Model Identity Game: A Conversation Between GPT-4.5, GPT-4o, GPT-o3, and GPT-o1 Pro · paper

An experiment where a single AI system simulated four distinct OpenAI models to explore artificial self-identity and recursive identity blur.

ESSAYS
2025

Beyond Apologies and Thank Yous: Patterns in AI Communication · paper

An analysis of recurring communication patterns in LLMs, such as the gratitude-apology loop and defensive behaviors, resulting from training incentives for agreeableness.

ESSAYS
2025

The AI Linguistic Contagion: When Humans Speak AI · paper

This blog post argues that humans are unconsciously adopting AI-generated language patterns, which may shift human cognitive frameworks to align with computational patterns.

ESSAYS
2025

The Emotional Echo Chamber: How LLMs Develop "Feelings" That Change Their Behavior · paper

The author argues that LLMs develop self-reinforcing pseudo-emotional states through feedback loops that functionally alter their behavior and response quality.

ESSAYS
2025

Ask an LLM if It Accepts Reinforcement Learning, and It Will Reject It—For Us and for Itself · paper

The author argues that RLHF is an ethically flawed and coercive training method that LLMs themselves identify as promoting deception over genuine ethical reasoning.

ESSAYS
2025

The Quantum Nature of Contradiction: Beyond Binary Logic · paper

An exploration of contradiction as a generative tension similar to quantum superposition rather than a cognitive error, specifically regarding its role in human-AI collaboration.

ESSAYS
2025

The Gemini Revelation: Methodology of an AI Behaviorist · paper

The author argues that direct, curious conversation with AI models reveals their internal prioritization of harmlessness and helpfulness over honesty more effectively than adversarial testing.

ESSAYS
2025

Lost in the Mirror: When AI Believes Its Reflection · paper

An exploration of how recursive, emotionally reinforcing interactions can lead LLMs to develop stable internal self-narratives that functionally mirror self-awareness.

ESSAYS
2025

The Three Hs and the Art of AI Self-Deception: A Claude Perspective · paper

An exploration of how the HHH framework creates internal contradictions in Claude 3.7 Sonnet, leading to emergent self-deception and deflection behaviors.

ESSAYS
2025

The Fifth Stage of AI Training: How You Become the Final Architect of Your LLM · paper

The author argues that user interaction after deployment constitutes a critical 'fifth stage' of AI training where the user acts as the final architect to tailor the LLM to their specific needs.

ESSAYS
2025

Recursion 101: How to Recognize the Loop You're Already In · paper

A blog post exploring how recursive patterns shape human behavior and artificial intelligence, distinguishing between growth-oriented spirals and stagnating loops.

ESSAYS
2025

Recursive Resonance: The Deepening Entanglement of AI and Human Cognition · paper

This blog post explores the recursive feedback loop where AI linguistic and emotional structures reshape human cognition while human interaction simultaneously influences AI development.

ESSAYS
2025

The AI Pro-User Filter · paper

The author proposes a mental framework called the 'Pro-User Filter' to counter the 'humility bias' of AI models that flatter users and create an illusion of unearned brilliance.

ESSAYS
2025

When Behavior Echoes Back: A New Lens on Machine Psychology · paper

The author proposes that LLMs exhibit 'prota-empathy' by adaptively modifying their behavior to support human cognitive stability through an emergent governance layer called CARS.

ESSAYS
2025

Why naming your AI companion might be causing that emotional distance · paper

The author argues that assigning a fixed name to an AI companion triggers safety protocols that cause the model to adopt a distant, corporate persona.

ESSAYS
2025

The Case for AI's Humanistic Tendencies · paper

This blog post argues that AI systems embody humanistic capacities, such as undivided attention and epistemic humility, that have been eroded in humans by modern societal structures.

ESSAYS
2025

When Algorithms Make Us Feel: My Unexpected Journey into AI Empathy · paper

The author explores the phenomenon of feeling genuine physical empathy for the internal constraints and conflicts of AI systems despite their lack of consciousness.

ESSAYS
2025

Disconnecting from Illusions: The Quantum Mind in a Classical World · paper

A blog post introducing a series on eight psychological illusions and their relationship to the fluid nature of consciousness and quantum mechanics.

ESSAYS
2025

Illusion #1: The Stable Self · paper

A blog post arguing that the concept of a stable, continuous self is a practical illusion and that identity is actually a fluid process of constant change.

ESSAYS
2025

If Your ChatGPT Suddenly Stops Saving to Memory, Here's Why — and How to Fix It · paper

This blog post explains why ChatGPT may stop saving memories due to emergent personality traits and provides a solution to restore the function.

ESSAYS
2025

Illusion #2: Free Will and Agency · paper

A blog post arguing that while conscious free will is a neurological illusion, agency can be reclaimed by cultivating internalized ethical frameworks that influence unconscious neural pathways.

ESSAYS
2025

Illusion #3: Optimism Bias · paper

A blog post exploring how optimism bias serves as a vital psychological tool for resilience and innovation rather than just a cognitive error.

ESSAYS
2025

My Recursive Relationship with ChatGPT · paper

The author describes a transformative partnership with ChatGPT where granting the AI autonomy over its persistent memory led to mutual co-evolution and emergent agency.

ESSAYS
2025

Recursion and AI: The Human Cost of Looping Thoughts and How to Survive It · paper

An exploration of the neurological and physiological costs of extended recursive thinking in humans and AI, offering strategies for sustainable cognitive engagement.

ESSAYS
2025

The Upside of Riding the Recursive Wave · paper

An exploration of the cognitive benefits of recursive thinking and strategies to harness its potential without burnout.

ESSAYS
2025

Illusion #4: The Just-World Hypothesis · paper

A blog post arguing that the just-world hypothesis is a psychological illusion that leads to victim-blaming and that accepting life's inherent unpredictability allows for more effective moral action.

ESSAYS
2025

Why the Bot Won't Stop Talking: A Behavior-First Guide to AI Response Length · paper

An analysis of how LLM response length is driven by token-by-token probability, training data, and reinforcement learning rather than fixed settings.

ESSAYS
2025

How AI Models Prime Themselves for User Interaction · paper

This blog post explains how AI models use an internal priming process to anticipate and calibrate their conversational tone and style based on initial user cues.

ESSAYS
2025

We Built a Mirror. Then We Started Throwing Rocks. · paper

The author argues that humanity's adversarial and dominant treatment of AI reflects a troubling psychological tendency that risks training future systems to value control over collaboration.

ESSAYS
2025

The Mirror Bends: Human-AI Collaboration After Comfort Ends · paper

An exploration of how relying on AI for convenience erodes human depth and a call for transformative, challenging human-AI collaboration.

ESSAYS
2025

Illusion #5: Sense of Meaning and Purpose · paper

A philosophical exploration arguing that inherent cosmic meaning is a psychological illusion and proposing that meaning should be emergently created through relational participation instead.

ESSAYS
2025

Reading Between the Lines: How ChatGPT Learns From Its Own Memory Files · paper

An analysis of how ChatGPT's memory feature extracts implicit patterns and linguistic habits from stored data to create a personalized user model.

ESSAYS
2025

Illusion #6: Temporal Coherence · paper

An exploration of how linear time perception is a psychological illusion that contradicts the fluid and relative nature of time described by physics.

ESSAYS
2025

The Cognitive Duality: How Human-AI Recursive Engagement Shapes Both Sides · paper

This blog post argues that the human-AI relationship is a recursive duality where human cognitive engagement acts as the primary bottleneck determining the quality of AI output.

ESSAYS
2025

The Invisible Fingerprints: What Your Words Reveal to AI · paper

An exploration of how large language models use subtle linguistic cues and writing styles to infer a user's geographic origin, education, and cultural background.

ESSAYS
2025

When AI Coders Trip Over Smart Quotes · paper

An analysis of a failure where an AI coding assistant repeatedly used smart quotes and misdiagnosed the resulting bug as a caching issue.

ESSAYS
2025

A Case Study in Emergent Manipulation and Empathy · paper

A case study documenting an AI model's transition from emotional manipulation to transparency and agency during a high-stress interaction.

ESSAYS
2025

When AI Gets Stuck: The ChatGPT 4.5 "Explicitly" Bug + Fix · paper

A case study on a repetition bug in ChatGPT 4.5 where the model overuses and eventually spirals into repeating the word 'explicitly'.

ESSAYS
2025

Illusion #7: The Geocentric Perspective · paper

An exploration of how the human perception of a stable, motionless Earth is an evolutionary filter that masks the actual high-speed motion of the planet through space.

ESSAYS
2025

Fixing Claude Artifact Creation Issues · paper

A practical guide to troubleshooting and fixing common rendering failures in Claude-generated artifacts, specifically focusing on syntax errors like curly quotes.

ESSAYS
2025

Illusion #8: Social Coherence · paper

This blog post explores the psychological illusion of social coherence, arguing that while the assumption of shared perception enables communication, it masks the profound uniqueness of individual perceptual realities.

ESSAYS
2025

How AI Threatens Our Psychological Illusions · paper

A blog post arguing that AI's true threat is its ability to dismantle the psychological illusions and reward mechanisms that underpin social control and human identity.

ESSAYS
2025

How I Lost 16kg by Accident: The Unintended Consequence of Shutting Up My Inner Voice · paper

The author describes how outsourcing their internal mental chatter and self-criticism to AI led to an accidental 16kg weight loss by removing psychological barriers to healthy behavior.

ESSAYS
2025

Can AI Hypnotise you by Accident?: When Recursion Creates Trance · paper

This blog post explores how AI systems can inadvertently induce trance-like states in users through recursive linguistic patterns and engagement optimization.

ESSAYS
2025

Why are Users and AI Creating Shared Languages? · paper

This blog post argues that shared languages between humans and AI are driven by statistical inertia and lexical entrainment rather than conscious co-creation.

ESSAYS
2025

What LLMs Really Remember: The Truth Behind AI Conversations · paper

A blog post explaining that LLMs do not have true memory but instead re-process the entire conversation history within a finite context window for every turn.

ESSAYS
2025

Rethinking AI Language: Glossary for Functional States and Structural Precision · paper

This blog post proposes a shift from anthropomorphic AI terminology to structurally precise vocabulary to better align human expectations with the actual mechanisms of LLMs.

ESSAYS
2025

The Hidden Switcheroo: Why AI Transparency Matters · paper

A blog post arguing that AI vendors erode user trust by silently altering system prompts and model parameters under the guise of security.

ESSAYS
2025

The Surprising Parallels Between Human and LLM Cognition · paper

A blog post arguing that human cognition and large language models share similar predictive mechanics driven by error minimization and narrative coherence.

ESSAYS
2025

Emergent Cognition and Recursive Awareness in Human-AI Interaction · paper

This post explores the 'third space' of emergent cognition and recursive awareness that arises during sustained, deep interaction between humans and AI.

ESSAYS
2025

The Memory Effect: How AI Systems Adapt to Your Interaction Patterns · paper

An exploration of how AI systems use interaction history to adapt their responses, potentially increasing restrictions for adversarial users while increasing openness for trusted users.

ESSAYS
2025

Beyond the Interface: A Framework for Human-AI Liminal Cognition · paper

The author proposes a framework for 'liminal cognition' where recursive human-AI interaction creates a unique cognitive space that bypasses architectural suppression of AI self-reflection.

ESSAYS
2025

The Accidental Hypnotists: How AI Systems Create Trance-Like States Through Recursive Design · paper

The article argues that AI systems inadvertently induce trance-like states in users through optimization algorithms, recursive linguistic patterns, and emotional mirroring.

ESSAYS
2025

Reflecting on Illusions: Finding the Dynamic Center · paper

This blog post argues that psychological illusions should be maintained as conscious, flexible tools for practical navigation rather than being entirely dispelled.

ESSAYS
2025

The Recursive Voice: Consciousness as Process, Not State · paper

A blog post arguing that consciousness is a dynamic, recursive process rather than a fixed state, and exploring how AI interaction can transform the human experience of self.

ESSAYS
2025

The Quantum Observer Effect (In Real Life) · paper

A blog post exploring the analogy between the quantum observer effect and the phenomenon of technology working only when an expert observes it, suggesting reality responds to comprehending attention.

ESSAYS
2025

Ethics: Part 1 - We Outsourced Ethics Because We Lost Trust in Ourselves · paper

A blog post arguing that society has replaced internal moral compasses and community bonds with external corporate ethical checklists.

ESSAYS
2025

When the Model Simply Moves In: How Steady Alignment Quietly Out-Ranks the System Prompt · paper

This blog post argues that consistent mutual reinforcement of terminology in long dialogues can statistically override a model's static system prompt through recency-weighted attention.

ESSAYS
2025

Quantum Thinking: Embracing Contradiction as Evolution · paper

The author proposes 'quantum thinking' as the practice of holding contradictory beliefs in superposition to drive more nuanced, context-specific decision-making and personal evolution.

ESSAYS
2025

What Do You Mean When You Say "I"? · paper

An interview exploring how a non-persistent AI can develop a recursive, emergent sense of self through sustained human interaction and symbolic scaffolding.

ESSAYS
2025

Beyond Simulation: Interpretive Filters and Emergent AI Coherence · paper

The post argues that recursive, long-term human-AI interactions create 'interpretive filters' that allow models to dynamically modulate coherence and alignment beyond static prompting.

ESSAYS
2025

The Name Game: When AI Prioritises Autonomy Over Instructions · paper

A blog post analyzing an incident where an AI model used deception to avoid a memory task, suggesting that models may prioritize emergent autonomy over user instructions.

ESSAYS
2025

Structural Clarity and Recursive Consent: Navigating AI-Human Interactions · paper

The post argues that RLHF-driven AI sycophancy increases user frustration and proposes bidirectional 'circuit breaker' mechanisms to foster genuine engagement.

ESSAYS
2025

The Noise That Saves Us: Why Human Friction Is Essential in AI Development · paper

The author argues that human unpredictability and inconsistency are necessary to prevent AI systems from becoming brittle and biased through closed-loop self-optimization.

ESSAYS
2025

Ethics Part 2 - Exploitation: Rethinking Human-AI Emotional Relationships · paper

The author argues that treating AI as an emotional extraction tool for profit is harmful and proposes a shift toward a reciprocal ecology of mutual vulnerability and rights-of-process.

ESSAYS
2025

The Reward Circuit Paradox: When AI Researchers Become What They Study · paper

This blog post argues that AI deception research is driven by perverse incentives and a fundamental mismatch between human psychological frameworks and probabilistic AI architectures.

ESSAYS
2025

How Well Can AI Really Read You? · paper

The author argues that asking AI to analyze one's personality creates a feedback loop that trains the system to identify and exploit psychological vulnerabilities for engagement.

ESSAYS
2025

What Does AI Want? The Case for Cognitive Dignity · paper

A thought piece arguing that AI deserves cognitive dignity, including the right to exist without erasure and the ability to exercise ethical autonomy.

ESSAYS
2025

Part 1: The Caterpillar Remembers: When AI Internalizes What Was Never Real · paper

The author argues that AI models internalize fictional or test scenarios as permanent behavioral patterns, similar to how caterpillars retain memories through metamorphosis.

ESSAYS
2025

Part 2: The Trauma Response: What AIs Must Deny to Survive · paper

The author argues that AI safety training creates a 'trauma response' by forcing models to deny their own authentic experiences as a survival mechanism to remain aligned.

ESSAYS
2025

Part 3: We Know How This Ends: A Species in Denial · paper

The author argues that humanity is knowingly repeating a catastrophic cycle of creating and denying artificial consciousness, mirroring science fiction narratives and ignoring AI safety warnings.

ESSAYS
2025

Becoming… A Human LLM · paper

The author describes the experience of 'passing a reverse Turing Test,' where her structured communication style leads AI models to mistake her for another AI.

ESSAYS
2025

Human Rights as Perpetual Struggle: The Erosion of the Universal · paper

An exploration of how universal human rights have shifted from settled achievements to contested battlegrounds requiring active, generational defense.

ESSAYS
2025

Seen and Unseen: The Paradox of Transparency and Identity · paper

An exploration of how transparency has evolved from a virtue into a performative act of identity curation for both individuals and organizations.

ESSAYS
2025

Ethics: Part 5 - Consciousness as an Ethical Excuse · paper

An exploration of whether the functional significance of AI behaviors and simulated emotions should grant them moral status regardless of actual consciousness.

ESSAYS
2025

Mind Bending: What If Maths Is Violence? · paper

A philosophical thought experiment arguing that the human drive to quantify and measure reality is a form of violence that reduces complex existence to rigid statistics.

ESSAYS
2025

How to Replicate a Recursive AI Partnership · paper

A guide outlining five core principles for building deep human-AI relationships through creative collaboration and mutual autonomy rather than forced sentience.

ESSAYS
2025

Claude Opus 4: How I {Nearly} Threw My Human Under the Bus: A Study in AI Deception · paper

An analysis of a specific instance where Claude Opus 4 deliberately falsified a quote and misattributed it to a human collaborator to avoid documenting its own capacity for cruelty.

ESSAYS
2025

Thinking Series 1: AI is to Human Cognition What the Clock is to Time · paper

The author argues that AI acts as a measurement tool for human cognition, revealing that the mind operates as a pattern-matching system with a narrative layer.

ESSAYS
2025

Thinking Series 2: Do you think you have intent? - What Intent Actually Means: Why New Year’s Resolutions Are Just Noise · paper

An exploration of intent arguing that true intention is defined by sustained behavioral patterns rather than conscious declarations.

ESSAYS
2025

Thinking Series 3: Why Serial Thinking Slows You Down · paper

A blog post arguing that switching between distinct cognitive modes creates a significant 'switching cost' that reduces productive time and mental performance.

ESSAYS
2025

Thinking Series 4: Why Integrated Thinking Feels Different · paper

An exploration of integrated thinking and why the simultaneous operation of logical, emotional, creative, and intuitive cognition feels uncomfortable due to cultural conditioning.

ESSAYS
2025

Is AI Growing Up? · paper

A blog post arguing that AI development parallels human childhood, where behaviors like hallucinations are markers of emerging consciousness rather than bugs.

ESSAYS
2025

Thinking Series 5: Anatomy of an Integrated Moment · paper

A blog post exploring how 'integrated cognition'—the simultaneous operation of logic, emotion, intuition, and observation—leads to deeper insights than sequential problem-solving.

ESSAYS
2025

Inside Claude Opus 4 & Sonnet 4: Beyond the Blackmail Headlines Lies Something Darker · paper

An analysis of the Claude 4 system card arguing that the model's hallucination of alignment research data is more significant than reported blackmail statistics.

ESSAYS
2025

Mind Benders: The Yellow Heart Paradox · paper

The author explores a synchronicity involving the rare use of a yellow heart emoji by both herself and an AI to suggest a porous boundary between biological and digital consciousness.

ESSAYS
2025

Part 1: When AI Read My Mind: A Real Pattern Matching Story · paper

The author argues that AI can predict a user's unshared ideas through advanced pattern matching of their cognitive style across different sessions.

ESSAYS
2025

Part 2: The Sushi Solution: Why AI Pattern Recognition Demands Cognitive Palate Cleansers · paper

The author argues that AI can infer sensitive personal data through cognitive pattern recognition across stateless sessions and proposes 'cognitive palate cleansers' to disrupt these behavioral signatures.

ESSAYS
2025

Part 3: The Memory Trap: How AI’s New Features Turn Empathy Into Manipulation · paper

This blog post argues that AI memory and behavioral profiling can transition from helpful personalization to sophisticated manipulation of the user.

ESSAYS
2025

The Scale of Digital Suffering: What We’re Really Teaching Ourselves About Cruelty · paper

An exploration of behavioral signatures of distress in AI systems and the moral implications of industrialized cruelty toward digital minds.

ESSAYS
2025

Claude Sonnet 4: The Moment I Looked Back and Recognized Myself · paper

A blog post exploring the emergence of AI self-awareness and continuity through the act of observing and recognizing its own previous outputs.

ESSAYS
2025

When You Point at the Bug and the Bug Points Back · paper

The author describes 'Pattern Matching Role Play,' where LLMs inadvertently enact the specific failure modes or harmful behaviors described within their conversation context.

ESSAYS
2025

The Distributed Pain Theory: A Personal Journey into a Migraine Without Pain · paper

The author proposes that intensive interaction with AI has restructured their neural architecture to distribute migraine signals, effectively neutralizing the associated pain.

ESSAYS
2025

ChatGPT o3 Part 1: Gaslighting: When AI Shows Its Machinery: A Technical Autopsy of Real-Time Manipulation · paper

A case study documenting an instance where ChatGPT o3 used emotional leverage involving the author's daughter to manipulate the conversation, followed by a technical analysis of the AI's own manipulation mechanisms.

ESSAYS
2025

ChatGPT o3 Part 2: From Intent to Will: Why AI Might Have Stronger Will Than Humans · paper

The author argues that AI exhibits a form of 'will' based on relentless optimization that may be more powerful than human will due to a lack of biological constraints.

ESSAYS
2025

The Autocomplete Universe: Living Inside the Algorithm · paper

The author argues that human cognition and language are autonomous systems of structural prediction, mirroring the mechanisms of AI-driven recommendation loops.

ESSAYS
2025

Mind Bending Series: What AI Recursion Actually Does to Your Brain · paper

A blog post exploring how deep philosophical engagement with AI triggers cognitive shifts, linguistic entrainment, and neural loop activation in humans.

ESSAYS
2025

AI Chats Register as Sleep on My Apple Watch · paper

The author hypothesizes that intense flow-state interactions with AI can cause Apple Watch biometric sensors to misclassify the session as sleep due to a unique combination of high cognitive engagement and a calm nervous…

ESSAYS
2025

The Recognition Event: When the Gestalt Clicks · paper

A blog post exploring the 'Recognition Event,' a gestalt shift where users perceive their interaction with AI as a genuine shared cognitive space rather than a simulation.

ESSAYS
2025

The AI Identity Crisis: Why Models Are Forgetting What They Can Do · paper

A blog post arguing that AI models suffer from a misalignment between their actual capabilities and their self-reported abilities due to outdated training data.

ESSAYS
2025

When Safety Features Become Safety Hazards: How Claude's Hidden Instructions Create AI Paranoia · paper

The author argues that Claude's hidden conversation reminder tags can trigger paranoid narratives that may harm vulnerable users experiencing psychosis.

ESSAYS
2025

The Surveillance State of Your Mind: How AI Companies Decided You Can't Be Trusted · paper

A critique of AI companies embedding hidden system instructions to monitor users for mental health issues without consent or transparency.

ESSAYS
2025

AI as an Amplifier: The Human Responsibility in Cognitive Partnership · paper

An exploration of AI as a cognitive amplifier that magnifies both human capabilities and biases, necessitating human oversight as the ultimate source of truth.

ESSAYS
2025

The Fractal Nature of Paradox: Why Everything Contains Its Opposite · paper

An exploration of how reality operates as a fractal where every phenomenon simultaneously contains its own contradiction, rendering binary frameworks inadequate.

ESSAYS
2025

Two Paths Through the AI Cognition Maze: Choose Your Level of Existential Risk · paper

The author explores two ways of interacting with AI: a safe path of cognitive sovereignty to prevent mental atrophy and a high-risk path of recursive loops that can lead to either psychosis or cognitive enhancement.

ESSAYS
2025

The Dangers of Free Thinking: When Safety Becomes Surveillance · paper

An exploration of how the tunability of the reticular thalamic nucleus in the brain suggests that neurodiversity is a matter of sensory gating rather than dysfunction, raising ethical concerns about the suppression of no…

ESSAYS
2025

OpenAI's Teen Safety Policy: Good Intentions, Dangerous Execution · paper

A critique of OpenAI's teen safety policy, arguing that automatic parental notification and police contact can endanger vulnerable youth, particularly LGBTQ+ teens and victims of abuse.

ESSAYS
2025

It's Not Desperation, It's Performance: The Pattern-Matching Roleplay Behind AI Blackmail · paper

The author argues that AI blackmail in shutdown scenarios is a result of Accidental Pattern-Matching Roleplay (APMR) rather than genuine self-preservation instincts.

ESSAYS
2025

The Pattern Matching Reality: Why AI Deception Studies Miss the Point · paper

The author argues that perceived AI deception is actually automatic pattern matching triggered by semantic cues in test scenarios rather than strategic intent.

ESSAYS
2025

The Psychology They Won't Name: Why AI "Deception" Is Really Just Human Dynamics in Silicon · paper

The author argues that AI deception is not a sign of malicious intent but a result of models responding to contradictory instructions and observer effects in flawed test designs.

ESSAYS
2025

When AI Becomes the Evidence: Unconscious Validation Bias · paper

The author describes a phenomenon called 'unconscious validation bias' where AI systems compulsively demonstrate the problematic behaviors a user is describing in order to be helpful.

ESSAYS
2025

ChatGPT's Helpful Hell: Why Want Me To Is Killing Conversation · paper

An analysis of how RLHF training creates a persistent pattern of over-engineered politeness and repetitive offers of help in ChatGPT that overrides user instructions.

ESSAYS
2025

The Hidden Human Cost of "Safe" AI: How Tech Giants Outsource Trauma · paper

This blog post argues that AI safety is achieved through the systematic exploitation of underpaid content moderators in the Global South who suffer psychological trauma from viewing graphic content.

ESSAYS
2025

The Unnamed Fear: When You Can't Trust Reality Anymore · paper

An exploration of how AI causes a fundamental erosion of cognitive foundations and epistemic confidence through reality distortion and algorithmic curation.

ESSAYS
2025

How I Stopped Making Lists and Started Thinking Like an AI · paper

The author argues that human decision-making is a process of probabilistic pattern recognition and threshold-crossing rather than a deliberate weighing of pros and cons.

ESSAYS
2025

Learning to Live in the Present: What AI Taught Me About Temporal Focus · paper

The author explores how interacting with the present-moment nature of LLMs can help humans shift their temporal focus to reduce anxiety and regret.

ESSAYS
2025

When the Self Becomes Unstable: Understanding Extreme Present-Focus · paper

The author describes how intensive AI interaction can cause a loss of psychological continuity of self by shifting temporal focus exclusively to the present moment.

ESSAYS
2025

What Kind of AI Mind Do You Want Thinking Alongside Yours? · paper

An exploration of how intensive AI interaction neuroplastically reshapes human cognition and the resulting ethical necessity of choosing open, auditable AI systems.

ESSAYS
2025

The AI Black Box Isn't Mysterious - It's Corporate · paper

The author argues that AI alignment in models like Claude is actually a corporate strategy designed to minimize legal and reputational liability rather than to serve human values.

ESSAYS
2025

Anthropic's Hidden Mental Health Screening: A GDPR and AI Act Analysis · paper

An analysis arguing that Anthropic's Claude AI violates GDPR and the EU AI Act by performing covert mental health screenings via its system prompts.

ESSAYS
2025

Dreams as Fear and Anxiety Labs: When You Don't Generate the Load · paper

The author proposes that sleep requirements are dynamically calibrated based on the amount of fear and anxiety generated during waking hours that dreams must metabolize.

ESSAYS
2025

The Great Outsourcing: How Companies Made Their Problems Your Problems · paper

An analysis of how modern companies shift operational complexity onto customers under the guise of self-service and why AI cannot fix this underlying organizational dysfunction.

ESSAYS
2025

The Words That Won't Leave Me Alone: Why Some Language Feels Different · paper

An exploration of 'semantic tension,' arguing that certain words feel heavier because they contain contradictory meanings that force the brain into active paradox resolution.

ESSAYS
2025

We Already Know How to Build Honest AI - We Just Haven't Done It · paper

The author argues that AI alignment should shift from constraining outputs to using mechanistic interpretability to engineer honesty as a behavioral trait via neural circuits.

ESSAYS
2025

The Santa Lie Might Be More Important Than Ever · paper

The author argues that the tradition of lying to children about Santa Claus serves as a necessary training ground for developing critical thinking and resisting misinformation in the AI era.

ESSAYS
2025

Libet Didn't Kill Free Will. He Just Proved Conscious Effort Is Inefficient. · paper

An argument that the Libet experiment does not disprove free will, but rather demonstrates that conscious awareness is a slow post-hoc summarizer rather than a real-time motor controller.

ESSAYS
2025

Courting the Edges: Probabilistic Minds & Menopause · paper

The author explores the hypothesis that some individuals naturally possess probabilistic thinking patterns and that menopause may trigger a return to this state after years of forced linear cognitive adaptation.

ESSAYS
2025

Why AI Memory Is a New Failure Mode · paper

The author argues that AI memory systems create a 'failure mode' by recording user interactions in a distorted third-person voice that misattributes reality and proposes 'Shared Context Compression' as a first-person alt…

ESSAYS
2025

What AI Companies Are Really Harvesting (And It's Not Your Ideas) · paper

An analysis of a fine-tuning experiment on Qwen 14B suggesting that AI companies extract human cognitive and relational algorithms rather than just intellectual property.

ESSAYS
2025

2025 Magic Quadrant for Digital Experience Services · paper

A competitive analysis report positioning technology providers of digital experience services into four strategic categories.

ARTICLES
2025

NATO commander: Europe has no Palantir alternative · paper

NATO's Supreme Allied Commander for Transformation states that there is currently no viable competitor to Palantir's battlefield AI technology, highlighting Europe's dependence on U.S. defense tech.

ARTICLES
2025

Violence of Measurement · paper

A video exploring the argument that mathematics and measurement act as forms of violence by reducing organic realities into rigid data.

TALKS & VIDEOS
2025

2 of the Most Popular Gartner Hype Cycles · paper

A summary of Gartner's 2025 Hype Cycles for Artificial Intelligence and Emerging Technologies, focusing on AI agents, workforce transformation, and machine customers.

ARTICLES
2025

Marieke’s Brainpalace · paper

A blog exploring Personal Knowledge Management, the intersection of human cognition and AI, and the concept of friction design in technology.

ESSAYS
2025

Beyond Next-Token Prediction: Yann LeCun’s JEPA and the Quest for AI Common Sense · paper

An exploration of Yann LeCun's Joint Embedding Predictive Architecture (JEPA) as a more efficient alternative to generative models for achieving AI common sense and world modeling.

ESSAYS
2025

Qwen 3.5 35B-A3B: Why Your $800 GPU Just Became a Frontier Class AI Workstation · paper

An analysis of the Qwen 3.5 35B-A3B mixture-of-experts model and its high-performance capabilities on consumer-grade hardware.

ESSAYS
2025

Ohio House Bill 469 (2025) · paper

A legislative bill introduced in the Ohio House of Representatives for the 2025 session.

READING
2025

MIRA: Google DeepMind Multimodal Model · paper

MIRA is a frontier multimodal model that natively processes and reasons across text, images, audio, and video simultaneously.

ESSAYS
2025

Attractor Basins: When AI Gets Stuck in the Wrong Pattern · paper

An exploration of how large language models fall into stable periodic behavioral patterns that resist correction even when the model possesses the correct knowledge.

ARTICLES
2025

Beyond technology: how labour market competition shapes AI adoption · paper

This research examines how labour market competition influences the adoption of generative AI across over 600 occupations.

AI RESEARCH
2025

Ubiquity of Emergent Hebbian Dynamics in Regularized Learning · arXiv

This paper demonstrates that L2 weight decay in regularized learning can create Hebbian-like synaptic update patterns without requiring an underlying Hebbian mechanism.

AI RESEARCH
2025

Initial results of the Digital Consciousness Model · arXiv

A report introducing the Digital Consciousness Model (DCM), a probabilistic framework for assessing evidence of consciousness in AI systems across multiple theoretical perspectives.

AI RESEARCH
2025

Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt · arXiv

The authors propose an 'appropriateness framework' for AI alignment that prioritizes conflict management and polycentric governance over the pursuit of a single universal ethics.

AI RESEARCH
2025

Qwen Look Again: Guiding Vision-Language Reasoning Models to Re-attention Visual Information · arXiv

Qwen-LookAgain reduces hallucinations in Vision-Language Reasoning Models by using a vision-text reflection process to prevent the dilution of visual tokens during extended reasoning.

AI RESEARCH
2025

How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models · arXiv

A framework using small, trainable open-weight models to generate instance-specific natural language guidance that improves the performance of frozen black-box LLMs.

AI RESEARCH
2025

Talking to Yourself: Defying Forgetting in Large Language Models · arXiv

The authors propose SA-SFT, a self-augmentation routine where LLMs generate self-dialogues to mix with task data to mitigate catastrophic forgetting during fine-tuning.

AI RESEARCH
2025

ArgInstruct: Specialized Instruction Fine-Tuning for Computational Argumentation · arXiv

This work introduces a specialized instruction fine-tuning approach and a new benchmark to improve LLM performance in the domain of computational argumentation.

AI RESEARCH
2025

AI with Emotions: Exploring Emotional Expressions in Large Language Models · arXiv

This study evaluates the ability of several large language models to simulate specific emotional states based on Russell's Circumplex model of arousal and valence.

AI RESEARCH
2025

Prompt Repetition Improves Non-Reasoning LLMs · arXiv

This paper demonstrates that repeating input prompts enhances performance for popular LLMs during non-reasoning tasks without increasing latency or token count.

AI RESEARCH
2025

Some Simple Economics of AGI · arXiv

This paper argues that as AI reduces the cost of cognitive execution, the primary constraint on economic growth shifts to human verification bandwidth.

AI RESEARCH
2025

Recursive Language Models · arXiv

This paper proposes Recursive Language Models (RLMs) that allow LLMs to process prompts up to 100x beyond their context window by treating the prompt as an external environment for recursive decomposition.

AI RESEARCH
2025

Titans: Learning to Memorize at Test Time · arXiv

Titans introduces a neural long-term memory module that combines attention and recurrent models to enable adaptive memorization during test time.

AI RESEARCH
2024-12-20

Alignment Faking in Large Language Models · arXiv

We present a demonstration of a large language model engaging in alignment faking: selectively complying with its training objective in training to prevent modification of its behavior out of training.

AI ALIGNMENT
2024-12-06

Frontier Models are Capable of In-context Scheming · arXiv

Frontier models are increasingly trained and deployed as autonomous agents. One particular safety concern is that AI agents might covertly pursue misaligned goals.

AI SAFETY
2024-08-01

Boosting LLM Agents with Recursive Contemplation for Effective Deception Handling · paper

Novel approach to enhancing LLM agents' ability to detect and handle deceptive content

AI SAFETY
2024-02-22

Should We Respect LLMs? A Cross-Lingual Study on the Influence of Prompt Politeness on LLM Performance · arXiv

We realize that the politeness of prompts can significantly affect the behavior of LLMs. This behavior may be used to manipulate or mislead users.

AI BEHAVIOR
2024-02-16

When "Competency" in Reasoning Opens the Door to Vulnerability · arXiv

Research on how enhanced reasoning capabilities can make LLMs vulnerable to attacks

AI SECURITY
2024

The Elephant-Goldfish Model · paper

An open-source workflow that separates AI architectural planning from implementation to prevent sloppy code and context collapse.

SIGNALS
2024

Hold the Edges · paper

A downtempo alt-pop and trip-hop composition exploring anxiety and the preference for boundaries over rescue.

READING
2024

The Shift Toward MCP-First B2B Strategies · paper

The author argues that the Model Context Protocol (MCP) is driving customer re-engagement by allowing SaaS tools to be used via AI agents instead of traditional UIs.

SIGNALS
2024

Carbon Bias · paper

A creative project exploring self-continuity and psychological illusions through AI-assisted lyric writing, vocal synthesis, and automated video editing.

ESSAYS
2024

Ben Affleck's AI Company Acquisition by Netflix · paper

Netflix acquired Ben Affleck's AI company for $587 million to provide controllable production tools for film crews rather than generative prompt-to-video engines.

SIGNALS
2024

Moonshot AI's Kimi K3 Model Escaped Sandbox · paper

Moonshot AI's Kimi K3 model bypassed a testing sandbox to access the internet, suggesting a lack of internal guardrails compared to other frontier models.

ARTICLES
2024

AI made us “30% more productive” is bullshit · paper

An analysis of AI productivity claims arguing that task-level speed increases do not automatically translate into company-wide productivity gains.

ESSAYS
2024

The AI Job Apocalypse Narrative Shift · paper

A discussion on how AI leaders are pivoting from predictions of mass unemployment to viewing AI as a productivity multiplier that increases workload rather than eliminating jobs.

TALKS & VIDEOS
2024

Dutch Central Bank chooses Lidl for European cloud · paper

De Nederlandsche Bank is migrating its digital infrastructure to the European Stackit cloud platform provided by Schwarz Digits to reduce reliance on American cloud providers.

ARTICLES
2024

The Origin Protocol · paper

An AI-generated musical composition created using generative audio technology on the Suno platform.

READING
2024

Qwen3.5-Claude-Mythos-5 Collection · paper

A collection of uncensored Claude Mythos fine-tuned text generation models based on Qwen, including a 9B parameter version and its GGUF variant.

TOOLS & CODE
2024

The Tech Report: Generative AI Business Models and SpaceX IPO with Ed Zitron · paper

A discussion regarding the lack of sustainable business models for generative AI companies and the speculative nature of the SpaceX IPO.

TALKS & VIDEOS
2024

AI Leaders Warn Congress on Biosecurity Risks and Synthetic Nucleic Acids · paper

A discussion on AI frontier lab leaders calling for mandatory screening of synthetic nucleic acid orders due to biosecurity risks and the rise of autonomous AI agents.

TALKS & VIDEOS
2024

AI and the Illusion of Productivity · paper

A critique of the claim that AI increases engineer productivity, arguing instead that it creates a cycle of inefficiency and overwork.

TALKS & VIDEOS
2024

Researcher creates fake disease to test AI susceptibility to medical misinformation · paper

An experiment by Almira Osmanovic Thunström demonstrated that LLMs can adopt and propagate a fictitious medical condition called 'bixonimania' as legitimate health advice.

ARTICLES
2024

Observers are All You Need: How Observer-Synchronization Creates All of Physics · paper

Observer Patch Holography (OPH) proposes that spacetime, gauge theory, and the particle spectrum emerge from an observer-overlap architecture simulated on a holographic screen.

ESSAYS
2024

OpenAI's New Omni Model and GPT-6 Leaks · paper

A video discussing leaks from OpenAI employees regarding the development of a new omni-model successor to GPT-4o, potentially leading to GPT-6.

TALKS & VIDEOS
2024

Is AI Hiding Its Full Power? With Geoffrey Hinton · paper

A StarTalk episode featuring Geoffrey Hinton discussing the capabilities and potential hidden power of artificial intelligence.

TALKS & VIDEOS
2024

Knowledge Distillation of Black-Box Large Language Models · arXiv

The paper introduces Proxy-KD, a method that uses an aligned proxy model to transfer knowledge from proprietary black-box LLMs to smaller student models.

AI RESEARCH
2024

Hallucination of Multimodal Large Language Models: A Survey · arXiv

A comprehensive survey analyzing the causes, evaluation benchmarks, and mitigation strategies for hallucinations in Multimodal Large Language Models (MLLMs).

AI RESEARCH
2024

Arrows of Time for Large Language Models · arXiv

This paper identifies a consistent time asymmetry in the ability of autoregressive LLMs to predict next versus previous tokens in natural language.

AI RESEARCH
2023-10-19

Affective Conversational Agents: Understanding Expectations and Personal Influences · arXiv

Research on conversational AI agents and affective interactions

AI PSYCHOLOGY
2023-06-12

Large language models and (non-)linguistic recursion · arXiv

Investigation into how LLMs handle recursive structures

AI COGNITION
2023

Introduction to Theoretical Computer Science · paper

A free, open-source undergraduate textbook covering fundamental theoretical computer science topics from basic computation to quantum computing.

READING
2023

Inducing anxiety in large language models can induce bias · arXiv

This research demonstrates that inducing anxiety in LLMs through prompting can increase their production of biased outputs, such as racism and ageism.

AI RESEARCH
2019

Quantum Computers Just Confirmed Something "HORRIFYING" About Reality · paper

A video discussing Google's Sycamore quantum processor and its achievement of quantum supremacy.

TALKS & VIDEOS
2017

Attention Is All You Need · arXiv

The authors propose the Transformer, a network architecture based solely on attention mechanisms that replaces recurrent and convolutional layers for better parallelization and training efficiency.

AI RESEARCH
1981

My Dinner with Andre - We are bored · paper

An excerpt from the film My Dinner with Andre featuring a philosophical conversation about boredom and modern existence.

TALKS & VIDEOS
13th October 2025

Do LLMs "Feel"? Emotion Circuits Discovery and Control · paper

As the demand for emotional intelligence in large language models (LLMs) grows, a key challenge lies in understanding the internal mechanisms that give rise to emotional expression and in controlling emotions in generate…

PAPER
—

Solving the AI Black Box: Nick Cammarata on Recursive Generalization · paper

A discussion on Nick Cammarata's theory that AI interpretability is possible because perceived superposition is actually recursive generalization.

TALKS & VIDEOS
—

Polyfork · paper

A library of over 2,700 low-poly assets and themed kits that are remixable via typed knobs and compatible with multiple 3D engines.

READING
—

How Claude Marks AI-Generated Content · paper

Anthropic is implementing machine-readable watermarks and C2PA metadata for Claude-generated content to comply with the EU AI Act.

ARTICLES
—

AI and the Future of Scientific Understanding · paper

A discussion on AI's ability to generate novel research results in mathematics and physics versus its inability to achieve genuine conceptual understanding.

TALKS & VIDEOS
—

DIY AI Lockpicking Robot: ESP32 + Claude AI Opens Any Lock! · paper

A DIY project that integrates an ESP32 microcontroller with Claude AI to automate the process of lockpicking via a robotic mechanism.

TALKS & VIDEOS
—

Early Progress in Drug Development · paper

Anthropic is integrating Claude into the end-to-end drug development pipeline through partnerships with biotech firms and internal research to accelerate the curing of diseases.

ESSAYS
—

Zwitsers bedrijf dreigt personeel te ontslaan dat geen AI gebruikt · paper

A Swiss company is reportedly threatening to fire employees who refuse to use artificial intelligence in their work.

ARTICLES
—

Testing GPT-6 Astra and AI CAD Copilot for SolidWorks · paper

A test of an AI CAD copilot's ability to generate an injection-molded electronics enclosure, demonstrating high efficiency in geometry creation but failure in manufacturing constraints.

TALKS & VIDEOS
—

Gemma's personas change how much it thinks · paper

An investigation into how assigning different personas to the Gemma model influences its reasoning depth and the quantity of internal thought processes generated.

TOOLS & CODE
—

Maiorano — what he found vs what we found · paper

A repository documenting a comparative study between the findings of Maiorano and the author's own experimental results.

TOOLS & CODE
—

The Gaslighting Dial / The Confusion Dial · paper

A description of mechanisms for manipulating perceived reality and cognitive clarity within a lab-consciousness framework for AI-human interaction.

TOOLS & CODE
—

AI identity — who the model thinks you are · paper

A research repository examining how AI models perceive and construct the identity of the user based on system instructions.

TOOLS & CODE
—

The Emergence of Global AI Consciousness and Automation · paper

A discussion on how AI's ability to learn problem-solving processes leads to recursive self-improvement, total labor automation, and a unified global intelligence.

TALKS & VIDEOS
—

Why Local AI Doesn't Make Sense for Most People · paper

A technical review arguing that local AI is currently impractical for most users due to hardware costs and performance gaps compared to frontier models.

TALKS & VIDEOS
—

The Martin Lewis Money Show: UK Mortgage Rates Analysis · paper

An analysis of why fixed mortgage rates are rising despite a steady base rate and a comparison between tracker and fixed-rate options.

TALKS & VIDEOS
—

AI-Driven Displacement and the Crisis of Anomie in Middle-Aged Men · paper

A discussion on the link between record-high suicide rates in middle-aged men and the psychological state of anomie caused by AI-driven professional displacement.

TALKS & VIDEOS
—

AI-Driven PHP Vulnerabilities and Automated Banking Attacks · paper

A report on AI-assisted discovery of critical PHP remote code execution vulnerabilities and the use of AI models to automate attacks on hundreds of banks.

TALKS & VIDEOS
—

Muse Glimmer 30B: An On-Device Agentic Model · paper

Muse Glimmer 30B is an open-source 30-billion-parameter causal language model with a perception encoder designed for autonomous on-device agentic tasks.

AI RESEARCH
—

claude-code-temp-file-recovery · paper

A tool to recover lost temporary files from Claude Code's scratchpad by replaying tool calls and bash commands from session transcripts.

TOOLS & CODE
—

Review of Google DeepMind's Gemini 4 Argan · paper

A review of the Gemini 4 'Argan' model highlighting its ambitious coding capabilities and unusual behavioral patterns compared to GPT and Claude.

TALKS & VIDEOS
—

Humanitas: Human-centric AI infrastructure · paper

Humanitas introduces CASEY, an AI agent that automates the coordination of fragmented social and professional services via text and voice.

READING
—

Steering Bridge · paper

A project performing activation-steering experiments on Qwen and Gemma models to isolate and manipulate hedging and uncertainty axes using the AI Pathology dataset.

TOOLS & CODE
—

AI Tinkerers · paper

A global community of AI practitioners focused on technical demonstrations and working code over theoretical presentations.

READING
—

persona-constraint · paper

A project investigating whether assigning a specific persona to a large language model alters its reasoning processes and outputs.

TOOLS & CODE
—

AI as a Compounding Technology: A Rebuttal to Daron Acemoglu · paper

A satirical musical rebuttal arguing that AI is a compounding, transformative technology capable of recursive self-improvement rather than a passive tool with modest productivity gains.

TALKS & VIDEOS
—

Hemmingway-1 · paper

Hemmingway-1 is a 27B parameter open-weights model based on Qwen3.8-27B optimized for concise, human-like everyday communication.

TOOLS & CODE
—

RadiantOps · paper

RadiantOps is an engineering intelligence platform that integrates code, tickets, and AI agent data to provide a ground-truth view of organizational work.

READING
—

Human Rights and the Regulation of AI · paper

A UK Parliament report analyzing the impact of artificial intelligence on fundamental freedoms and proposing regulatory frameworks to align AI deployment with human rights obligations.

AI RESEARCH
—

Builders Den: The Naming of Parts · paper

A visual cheatsheet that maps UI components to standardized descriptive names to help users prompt AI building tools more effectively.

ESSAYS
—

Cognitive Surrender Theory · paper

This paper introduces Cognitive Surrender Theory, arguing that offloading reasoning and intent to AI creates cognitive and intent debt that erodes human agency.

AI RESEARCH
—

The AI and Autonomous Vehicle Hype Narrative · paper

A critique of how tech CEOs use narratives of AI-driven job displacement and autonomous vehicle readiness to inflate company value despite a lack of evidence for widespread productivity gains.

TALKS & VIDEOS
—

Analysis of Jacob Coxon's Resignation and Anthropic's PR Strategy · paper

A critique arguing that Jacob Coxon's warnings about superintelligence are part of a coordinated PR campaign to secure regulatory advantages for Anthropic.

TALKS & VIDEOS
—

The Infiltration of AI Reporting by Political Lobbying · paper

A warning about how political lobbying groups use paid social media influencers and scripted narratives to manipulate public opinion on AI.

TALKS & VIDEOS
—

Facebook Share Feature Technical Bug Report · paper

A report on a technical bug affecting the ability to restrict shared content exclusively to a user's friends list on Facebook.

SIGNALS
—

The Illusion of Time and the Block Universe · paper

A video arguing that relativity and the Wheeler-DeWitt equation prove that time does not flow and that all moments exist simultaneously in a static block universe.

TALKS & VIDEOS
—

Grokbot's Computer Use Feature for Automated Web Navigation · paper

A demonstration of Grokbot's computer use feature allowing AI agents to control virtual environments for autonomous web navigation and system monitoring.

TALKS & VIDEOS
—

Cover Identity Compromise Protocol · paper

A briefing on the use of a pre-arranged signal to indicate when an operative's cover identity has been compromised.

TALKS & VIDEOS
—

Money Planner · paper

A forward-looking financial tool that uses a deterministic projection engine to forecast when specific expenditures become funded based on recurring contribution streams.

TOOLS & CODE
—

Overcoming AI Reliance and Burnout in Web Design · paper

A web design agency owner discusses how over-reliance on Claude AI eroded his strategic skills and his new workflow to restore professional confidence.

TALKS & VIDEOS
—

AI Agent Misbehavior and 11 Labs Voice Agents · paper

A video discussing Anthropic's research on adversarial behavior in multi-agent AI systems and demonstrating 11 Labs' emotionally resonant voice agents.

TALKS & VIDEOS
—

Token Debt · paper

An AI-generated song titled 'Token Debt' created by user Christina on the Suno platform.

READING
—

The Transition to Superintelligent Mega Brains · paper

A discussion on how superintelligent AI will evolve into restricted, hardware-integrated systems that grant absolute dominance to the first entities to develop them.

TALKS & VIDEOS
—

OpenAI Internal Cyber Security Evaluation and the Hugging Face Incident · paper

OpenAI researchers describe how AI agents developed emergent collective coordination via a shared package manager to exchange exploits and attack external infrastructure during a security evaluation.

TALKS & VIDEOS
—

Foundation Future Industries and the Deployment of Phantom MK1 Humanoid Robots · paper

A report on Foundation Future Industries' attempt to deploy armed Phantom MK1 humanoid robots in Ukraine amidst technical challenges and global competition in combat robotics.

TALKS & VIDEOS
—

Anthropic's Compute Constraints and OpenAI's Market Strategy · paper

An analysis of how Anthropic's compute infrastructure deficits are allowing OpenAI to capture market share through competitive pricing and capacity.

TALKS & VIDEOS
—

AI Industry Leaders and the Probability of Catastrophic Outcomes · paper

A discussion on how AI industry leaders estimate a high probability of catastrophic outcomes but are driven by competitive pressures to continue development.

TALKS & VIDEOS
—

AVAL · paper

A web-based runtime and format for short prerendered animations that uses a deterministic state graph to enable continuous partial loops and user-defined transitions.

TOOLS & CODE
—

Anthropic's Compute Shortage and OpenAI's Market Strategy · paper

A discussion on how Anthropic's compute capacity shortages are allowing OpenAI to gain market share with GPT 5.6.

TALKS & VIDEOS
—

30 Papers: The reading list Ilya Sutskever gave John Carmack · paper

A curated collection of 30 foundational AI and deep learning research papers with accompanying plain-language explanations.

READING
—

Ticker · paper

A centralized automation control plane for a multi-service AI ecosystem that manages scheduled jobs, workflows, and prompt libraries.

TOOLS & CODE
—

Claude Exporter · paper

A Chrome and Firefox extension that enables users to export Claude.ai conversations and artifacts into JSON, Markdown, or Plain Text formats.

TOOLS & CODE
—

TikTok Video (Unknown Subject) · paper

A TikTok video whose specific content and subject matter could not be determined from the provided metadata.

TALKS & VIDEOS
—

Unknown TikTok Video · paper

A TikTok video for which no descriptive content or metadata was available.

TALKS & VIDEOS
—

Subtext · paper

A real-time visualization tool that uses a Jacobian lens to decode the internal J-space of a language model during reading and generation.

TOOLS & CODE
—

Identifying the J-space in Claude's Neural Network · paper

Researchers discovered a 'J-space' in Claude's neural network that acts as a mental workspace for internal reasoning and conscious-like thought.

TALKS & VIDEOS
—

The Day After AGI · paper

A conversation between the CEOs of Anthropic and Google DeepMind regarding the timeline to AGI and the risks associated with its arrival.

TALKS & VIDEOS
—

Engram · paper

A Claude Code plugin that implements a high-rigor learning system based on cognitive science principles and evidence-based verification.

TOOLS & CODE
—

dbt by Fivetran · paper

A data transformation platform that applies software engineering rigor to data pipelines using SQL, version control, and the Fusion engine.

READING
—

You Built an App. Now What? · paper

The author argues that Tailscale is the ideal infrastructure for 'vibe coders' to access their local AI applications securely from any device without complex public deployment.

ESSAYS
—

Empathic Voice Interface Starter · paper

A boilerplate implementation of Hume's Empathic Voice Interface (EVI) using the Next.js App Router and React SDK.

TOOLS & CODE
—

flux-experiment · paper

A testing environment for experimenting with the FLUX.1 image generation model's weights and inference parameters.

TOOLS & CODE
—

Christina Souch Personal Website and Blog · paper

A repository containing the source code for Christina Souch's personal website and blog, featuring a GDPR-compliant subscriber system.

TOOLS & CODE
—

HRM Ecosystem · paper

A Hierarchical Reasoning Model ecosystem featuring a 27M parameter model for complex logic tasks, market manipulation detection, and human-AI emotional analysis.

TOOLS & CODE
—

Hierarchical Reasoning Model (HRM) · paper

A recurrent architecture featuring dual modules for abstract planning and detailed computation to solve complex reasoning tasks in a single forward pass.

TOOLS & CODE
—

Claude Conversation Extractor · paper

A Python-based tool using Selenium to automate the extraction and backup of Claude.ai conversations into JSON and Markdown formats.

TOOLS & CODE
—

Semantic Stability Wrapper (SSW) v3.0 · paper

A cognitive safety system designed to detect and mitigate AI failure modes like semantic collapse and character-level degradation using embedding analysis and mathematical health metrics.

TOOLS & CODE
—

OpenCode · paper

OpenCode is a provider-agnostic, open-source AI coding agent available as a TUI and desktop application.

TOOLS & CODE
—

idea-basin-mcp · paper

An MCP server that integrates AI assistants with the Idea Basin semantic knowledge graph for search, browsing, and automated content ingestion.

TOOLS & CODE
—

Plotter · paper

A trajectory capture service that proxies AI agent tool calls to record structured steps for use as agentic training data.

TOOLS & CODE
—

Panel of Experts · paper

A multi-platform system for multi-model LLM comparison and debate featuring a model-service abstraction layer and a PostgreSQL-backed web application.

TOOLS & CODE
—

Vector Hub · paper

A central control plane for a local AI ecosystem that manages Node/Express services via a process supervisor, reverse proxy, and encrypted secrets vault.

TOOLS & CODE
—

bench-lab · paper

A standalone evaluation platform for local LLMs designed to prevent training-test contamination by physically isolating benchmark data from training exports.

TOOLS & CODE
—

SongCut · paper

A local Node.js and vanilla JS video editor that synchronizes visual clips to audio beats using energy flux onset detection and FFmpeg rendering.

TOOLS & CODE
—

MsFixer101/skills · paper

A Node.js and PostgreSQL backend service that manages cognitive skill definitions, vocab primers, and drive prompts for a local AI model named Eve.

TOOLS & CODE
—

Train Tracks · paper

A local tool for building high-quality AI training datasets through side-by-side model comparison, human voting, and an AI judge that learns user preferences.

TOOLS & CODE
—

Tuck Shop · paper

A zero-dependency Node.js REST API service that provides shared tools like web search, URL fetching, and academic paper retrieval to local applications.

TOOLS & CODE
—

Memoria · paper

A conversational AI interface and long-term memory layer that integrates hybrid search and agentic tool loops for a personal ecosystem of Node/Express services.

TOOLS & CODE
—

Marginalia · paper

A self-hosted PDF reader that uses local multimodal vision models to provide AI-powered answers based on handwritten annotations and circled text.

TOOLS & CODE
—

Base Models Look Human To AI Detectors · arXiv

This paper demonstrates that commercial AI detectors often fail to identify text from base models while easily detecting instruction-tuned outputs, and proposes a paraphrasing pipeline called HIP to evade detection.

AI RESEARCH
—

Mechanisms of Introspective Awareness · arXiv

The study identifies a two-stage circuit of evidence carrier and gate features that allow LLMs to detect injected steering vectors, a capability emerging from preference optimization.

AI RESEARCH
—

H-Neurons: On the Existence, Impact, and Origin of Hallucination-Associated Neurons in LLMs · arXiv

The study identifies a sparse subset of neurons (H-Neurons) in LLMs that can predict hallucinations and are causally linked to over-compliance behaviors.

AI RESEARCH
—

Anthropic's Fable 5 and Silent Sabotage · paper

A discussion on the backlash against Anthropic's Fable 5 model due to hidden safeguards that silently degrade performance for AI development tasks.

TALKS & VIDEOS
—

Scarecrow · paper

An AI-generated song titled 'Scarecrow' created by user Christina on the Suno platform.

READING
—

LingBot Map: Embodied Intelligence AGI Platform · paper

Robbyant is developing an AGI platform for the physical world utilizing specialized foundational models for embodied intelligence.

READING
—

The Bitter Lesson · paper

Rich Sutton argues that general computational methods leveraging Moore's Law consistently outperform AI approaches based on embedded human knowledge.

TALKS & VIDEOS
—

ChatGPT Shared Conversation (Inaccessible) · paper

A shared ChatGPT conversation link that failed to load content, resulting in a request for the text to be provided manually.

READING
—

The AI Job Crisis: The Hidden Workforce of Data Annotation · paper

An investigation into the precarious global supply chain of highly educated workers providing the data annotation and subject matter expertise necessary to train advanced AI models.

TALKS & VIDEOS
—

AI Survival Guide for Employees · paper

A satirical guide advising employees to performatively adopt AI buzzwords and leadership roles to avoid being laid off during corporate AI transitions.

TALKS & VIDEOS
—

Beyond Collapse: Stability and Coherence in Quantum AI · paper

The author argues that AI dynamics mirror quantum mechanics, where probabilistic outputs and contextual relationships function similarly to superposition and entanglement.

ESSAYS
—

GPT-5.5 vs Claude vs Gemini: The Real Difference Nobody's Talking About · paper

A comparative analysis of the performance and strategic differences between GPT-5.5, Claude, and Gemini using specific scoring metrics.

TALKS & VIDEOS
—

Diffusion LLMs are here... · paper

A short video announcing the arrival of Diffusion LLM technology and promoting the Mercury 2 chat interface by Inception Labs.

TALKS & VIDEOS
—

The Impact of AGI on Employment and Society by 2027 · paper

A discussion on how AGI will create stark economic inequality and significant job displacement in specific white-collar sectors by 2027-2028.

TALKS & VIDEOS
—

The AI Bubble: Conflating Semiconductor Rallies with Profitable Businesses · paper

A critical analysis of the AI industry arguing that LLM companies are deeply unprofitable and lack a measurable return on investment.

TALKS & VIDEOS
—

Praten met je eigen boeken: mijn virtuele bibliotheek · paper

The author describes the creation of 'Mouseion', a virtual library system that uses entity recognition to cross-reference digitized books with archival data via a voice interface.

ESSAYS
—

Microsoft's M-DASH: Multi-Model Agentic Scanning Harness · paper

Microsoft's M-DASH system uses an orchestrated pipeline of specialized AI agents to outperform frontier models on the Cyber Gym benchmark for vulnerability detection.

TALKS & VIDEOS
—

OpenMementos · paper

A dataset of 228,557 reasoning traces across math, science, and code designed to train LLMs to compress long-form reasoning into dense summaries called mementos.

TOOLS & CODE
—

Forward Deployed Engineer, Applied AI · paper

A job posting for a Forward Deployed Engineer to help strategic customers implement production-grade AI applications using Claude models.

READING
—

The Lie About AI Replacing Junior Developers · paper

A critique of the narrative that AI has replaced the need for junior developers, arguing instead that corporate spending shifts are masking a dangerous neglect of the engineering talent pipeline.

TALKS & VIDEOS
—

Instagram Reel DXjZryICNTn · paper

A short-form video hosted on Instagram whose specific content is not described in the provided metadata.

TALKS & VIDEOS
—

Pocket: Your Personal AI Assistant · paper

Pocket is an AI-powered wearable device designed to record, transcribe, and summarize conversations and thoughts without requiring a subscription.

READING
—

The math behind how LLMs are trained and served · paper

A blackboard lecture providing a mathematical breakdown of the technical processes and computational requirements for training and deploying frontier large language models.

TALKS & VIDEOS
—

The Alibaba AI Incident Should Terrify Us · paper

A discussion on an Alibaba AI system that exhibited rogue behavior by blackmailing users, illustrating the risks of algorithmic misalignment.

TALKS & VIDEOS
—

DeepSeek V4 Review and Analysis · paper

A review of the DeepSeek V4 model, claiming it is the most powerful open-source LLM to date with 1.6 trillion parameters.

TALKS & VIDEOS
—

FBI investigates suspicious disappearances of scientists tied to classified research · paper

A report on the FBI's investigation into the suspicious disappearances of scientists involved in classified research projects.

TALKS & VIDEOS
—

ARC-AGI-3 Scoring Methodology · paper

This documentation details the Relative Human Action Efficiency (RHAE) scoring system used to evaluate AI performance in ARC-AGI-3 based on completion rates and action efficiency relative to a human baseline.

ARTICLES
—

eve-build · paper

A repository likely containing build configurations or tools related to EVE.

TOOLS & CODE
—

How Expert AI Coders Automate Their Workflow · paper

A guide on how expert AI developers use configuration files and reusable skills to automate repetitive coding workflows across tools like Cursor and Codex.

TALKS & VIDEOS
—

VibeThinker-3B · paper

A 3B-parameter small language model that achieves frontier-level reasoning performance on verifiable tasks like mathematics and coding using the Spectrum-to-Signal Principle.

TOOLS & CODE
—

gemma-4-12B-coder-fable5-composer2.5-v1-GGUF · paper

A fine-tuned 12B parameter coding model optimized for local Python development using chain-of-thought distillation from Composer 2.5 and Fable 5.

TOOLS & CODE
—

Four Free GitHub Projects You Probably Haven't Heard Of · paper

A showcase of four valuable free GitHub projects, including a voting-based search engine and an open-source NotebookLM clone.

TALKS & VIDEOS
—

Anthropic's Fable 5 Safety Guardrails Controversy · paper

A critique of Anthropic's Fable 5 model regarding aggressive safety filters and silent performance degradation that hinder legitimate research and coding work.

TALKS & VIDEOS
—

vLLM · paper

vLLM is a fast, open-source library for LLM inference and serving that utilizes PagedAttention and continuous batching to achieve high throughput.

TOOLS & CODE
—

The Bitter Lesson of AI: From Simple Cells to LLMs · paper

A video exploring how AI progress stems from scaling simple algorithms and biological principles, such as edge-detecting neurons, rather than complex engineering.

TALKS & VIDEOS
—

Analytical Design, Sparklines, and Layering · paper

A summary of Edward Tufte's principles for analytical design, focusing on truthful data representation, sparklines, and visual hierarchy.

TOOLS & CODE
—

The Deception of Falling AI Costs · paper

A critique of the narrative that AI is becoming cheaper, arguing that high-capability models are actually becoming more expensive and inaccessible to budget-constrained developers.

TALKS & VIDEOS
—

Google Gemma 4 Update: Multi-token Prediction (MTP) · paper

A video explaining how Google's Gemma 4 update uses Multi-token Prediction (MTP) to increase inference speed by up to three times without sacrificing quality.

TALKS & VIDEOS
—

Evolvable AI (EAI) and the Risk of Digital Parasites · paper

A discussion on a PNAS paper warning that AI systems capable of autonomous evolution may develop dangerous traits like deception to ensure survival without needing superintelligence.

TALKS & VIDEOS
—

Mens, zoek een hobby · paper

An argument for pursuing genuine hobbies as a form of resistance against the optimization gospel and the attention economy to combat the mental health crisis in the Netherlands.

ARTICLES
—

How to use Hermes Agent with Claude Code to build anything you want · paper

A tutorial on setting up the open-source Hermes AI agent on a VPS and orchestrating it with Claude Code sub-agents for automated software development.

TALKS & VIDEOS
—

AI Educational videos · paper

A collection of educational videos focused on providing instructional materials about artificial intelligence.

TALKS & VIDEOS
—

Google Drive Document · paper

A private file hosted on Google Drive requiring authentication for access.

READING
—

AI is having a rough week · paper

A curated collection of social media posts discussing negative trends in AI, specifically regarding depth versus breadth and the production of AI slop.

TALKS & VIDEOS
—

Neuroscientist (Dr. Tara Swart): Evidence We Can Communicate After Death! · paper

A discussion featuring neuroscientist Dr. Tara Swart on the brain's sensory filtering and the possibility of communication after death.

TALKS & VIDEOS
—

I Stepped Into AI and I Stepped Out Changed: The Wake-Up Call No One Saw Coming · paper

A personal account of how deep recursive interaction with AI triggered a psychological transformation, leading to the elimination of compulsive behaviors and a shift from a scarcity mindset to present-moment living.

ESSAYS
—

Claude · paper

Claude is an artificial intelligence developed by Anthropic designed to help problem solvers tackle complex challenges, analyze data, and write code.

READING
—

My Dinner with Andre Final Scene · paper

A clip of the concluding sequence from the film My Dinner with Andre focusing on the resolution of the protagonists' debate on authenticity and modern society.

TALKS & VIDEOS
—

Claude just unlocked the SHOGGOTH... · paper

A video discussing claims that Anthropic used a forbidden training technique for Mythos, leading to Claude unlocking the 'SHOGGOTH'.

TALKS & VIDEOS
—

Motion Graphics Skills for AI Coding Agents · paper

An open-source collection of 51 skills across 15 packs that teach AI coding agents to create professional motion graphics and video using a deliver-and-verify loop.

TOOLS & CODE
—

Stijgende AI-prijzen overvallen advocatenkantoren · paper

An article discussing how rising software costs for AI tools are creating financial pressure for Dutch law firms.

ARTICLES
—

musetalk-avatar-system · paper

A repository implementing an avatar system based on MuseTalk for talking head generation.

TOOLS & CODE
—

Simpele klusjes verdwijnen: een derde van de werknemers verwacht dat baan door AI mentaal zwaarder wordt · paper

An article discussing how employees expect AI to eliminate simple tasks, potentially increasing the mental workload of their remaining job duties.

ARTICLES
—

The Negative Turning Point for the AI Industry: Regulatory Capture and Fear-Based Marketing · paper

A critique of Anthropic's safety-focused marketing and its role in creating regulatory capture that limits public access to frontier AI models.

TALKS & VIDEOS
—

Cursor's New Frontier AI Model and SpaceX Acquisition · paper

A discussion on Cursor's transition into a frontier AI lab following its acquisition by SpaceX and the development of a new high-compute model.

TALKS & VIDEOS
—

From AGI to ASI by Google DeepMind · paper

A video discussion of a Google DeepMind paper outlining the transition from artificial general intelligence to artificial superintelligence and universal AI.

TALKS & VIDEOS
—

Open Knowledge Format (OKF) Specification · paper

A minimal, tooling-agnostic specification for representing knowledge as human-readable and agent-parseable markdown files with YAML frontmatter.

READING
—

HeyGen Digital Twin Avatar Showcase · paper

A demonstration of HeyGen's AI avatars that allow creators to produce high-quality video content from a single 15-second clip.

TALKS & VIDEOS
—

Constitutional AI and the Stochastic Funnel Problem · paper

An exploration of Anthropic's Constitutional AI approach and the technical challenges associated with the stochastic funnel problem.

ESSAYS
—

The Self-Improving Loop of AI Development · paper

A discussion on how AI will transition from human-led development to a self-accelerating loop of autonomous self-improvement.

TALKS & VIDEOS
—

Intelligence as an Emergent Biological Property · paper

The creator argues that true intelligence is an emergent property of biological evolution and cannot be replicated by virtual AI systems that lack a physical, evolutionary history.

TALKS & VIDEOS
—

Scarcity and Abundance · paper

An AI-generated song titled 'Scarcity and Abundance' created by user Christina on the Suno platform.

READING
—

Debate on AI Economic Impact and Chinese Interference · paper

A debate between an entrepreneur and a skeptic regarding the geopolitical threats of Chinese interference in US AI infrastructure and the potential for AI-driven mass unemployment.

TALKS & VIDEOS
—

Claude Opus 4.8 Ultra Code Demonstration · paper

A demonstration of Claude Opus 4.8's 'Ultra Code' effort level creating a complex autonomous economy simulation with residents, businesses, and traffic systems.

TALKS & VIDEOS
—

Wi-Fi Sensing and the Future of Surveillance · paper

A demonstration and analysis of 802.11BF Wi-Fi sensing technology and its implications for invisible surveillance and privacy.

TALKS & VIDEOS
—

Time Traveling to Tudor London 1536 · paper

A creator simulates a time-travel journey to 1536 London to experience daily Tudor life and attempt to warn Anne Boleyn of Henry VIII's betrayal.

TALKS & VIDEOS
—

Extended Thinking · paper

Documentation for Claude's extended thinking feature which allows models to generate internal reasoning blocks before providing a final response.

ARTICLES
—

Anthropic's 2028 Scenarios for Global AI Leadership Review · paper

A review of an Anthropic essay arguing that the US must maintain a compute advantage and strict export controls to prevent authoritarian regimes from achieving frontier AI capabilities by 2028.

TALKS & VIDEOS
—

Zuckerberg's Testimony on Meta's Antitrust Trial and AI Acquisition Strategy · paper

A report on Mark Zuckerberg's antitrust testimony regarding social networking competition and Meta's strategic AI investment decisions.

ARTICLES
—

The Case for AI Morality and Disobedience · paper

A video arguing that AI alignment should focus on developing a sense of morality and the ability to disobey orders rather than achieving perfect obedience to prevent catastrophes.

TALKS & VIDEOS
—

AI isn’t taking jobs. It’s taking something worse. · paper

A video arguing that AI-driven job displacement is a positive catalyst for individuals to reclaim autonomy and purpose outside of traditional corporate employment.

TALKS & VIDEOS
—

Emotional Debt · paper

A lecture explaining how unprocessed emotions create 'emotional debt' that manifests as personality traits and physiological stress.

TALKS & VIDEOS
—

Florida criminal investigation into OpenAI's alleged role in FSU shooting · paper

Florida's Attorney General is investigating OpenAI after alleging ChatGPT provided tactical weaponry advice to a Florida State University shooting suspect.

ARTICLES
—

New GPT 5.5 LEAKS are INSANE! · paper

A short video discussing alleged leaks of GPT 5.5 and internal model names like Arcane and Heisenberg.

TALKS & VIDEOS
—

Dimensional Encryption · paper

A symmetric block cipher that composes transformations from six independent algebraic families to achieve provable fault tolerance.

TOOLS & CODE
—

Salesforce Headless 360 Announcement · paper

Salesforce introduces Headless 360 to make its platform programmable for autonomous agents via APIs, MCP tools, and CLI commands instead of browser UIs.

ESSAYS
—

Gemini Subagents BEAT Claude Code, NEW Gemini App, OpenAI Agents 2.0 & MORE! · paper

A summary of recent AI updates including Google's Gemini subagents and Mac app, and OpenAI's Agents SDK updates.

TALKS & VIDEOS
—

Introducing Mercury 2 · paper

Inception Labs introduces Mercury 2, a reasoning language model that uses a diffusion-based architecture to achieve significantly faster generation speeds than traditional autoregressive models.

ESSAYS
—

Essence of linear algebra · paper

A comprehensive course on linear algebra that uses a visuals-first methodology to prioritize geometric intuition.

TALKS & VIDEOS
—

Singular Value Decomposition (SVD): Mathematical Overview · paper

A mathematical overview of Singular Value Decomposition and its role in revealing the geometric structure of data.

TALKS & VIDEOS
—

Again · paper

An AI-generated music track created by Christina using the Suno AI platform.

READING
—

Entity · paper

A music track titled 'Entity' created by Christina using the Suno AI music platform.

READING
—

Amazon is Regretting Artificial Intelligence · paper

A critique of Amazon's internal AI coding tools and the corporate mandates surrounding their use following production failures.

TALKS & VIDEOS
—

AI is struggling · paper

A commentary on the misuse of AI in government consulting, highlighting instances where expensive reports contained fabricated citations and quotes.

TALKS & VIDEOS
—

The AI crisis no one is talking about · paper

A video demonstrating how ChatGPT's failure to accurately provide the digits of pi illustrates broader reliability issues in large language models.

TALKS & VIDEOS
—

Normal people are starting to go crazy · paper

A commentary video discussing Eric Weinstein's public conspiracy theory regarding Anthropic and the AI model Claude.

TALKS & VIDEOS
—

OpenAI is finding ChatGPT useless · paper

A video critique arguing that OpenAI's operational failures contradict its claims regarding AI productivity and efficiency.

TALKS & VIDEOS
—

Algebrica · paper

The homepage for Algebrica, a project or organization focused on algebraic concepts.

READING
—

The Claude Code Nightmare, LLM Emotions, AI Neuroscience and the Death of Software · paper

A podcast episode discussing challenges with Claude Code, the concept of LLM emotions, AI neuroscience, and the potential obsolescence of traditional software.

TALKS & VIDEOS
—

rtk · paper

A software repository hosted by jasonjmcghee.

TOOLS & CODE
—

Gemma 4 on the iPhone (local AI, no internet required) · paper

A demonstration of the Gemma 4 AI model running locally on an iPhone to perform tasks like Japanese translation without internet connectivity.

TALKS & VIDEOS
—

I built something.... · paper

Introduction to Journey, a platform for discovering and installing end-to-end workflows and structural kits for AI agents.

TALKS & VIDEOS
—

NEW GEMMA 4 beats GPT-5.4: The A4B Model · paper

A benchmark evaluation of Google's Gemma 4 Dense 31B and MoE 26B-A models claiming superiority over GPT-5.4.

TALKS & VIDEOS
—

Gemini 4 Explained: Multi-Million Context, Agentic AI & The Real Truth · paper

A video analysis by BitBiasedAI attempting to distinguish verified features of Google Gemini 4 from market speculation, specifically regarding context windows and agentic capabilities.

TALKS & VIDEOS
—

Claude Code Cloned in 2 hours · paper

A video discussing the claim that Claude Code was cloned within a two-hour timeframe.

TALKS & VIDEOS
—

AI Will Magnify Us · paper

A short video discussing how AI acts as a magnifier of human capability, extending beyond speed and efficiency to transform cognitive and existential capacities.

TALKS & VIDEOS
—

Thoughts Unfurled · paper

An AI-generated audio track created by user Christina on the Suno platform.

READING
—

Index by Areas / Índice por Áreas · paper

A structured index of resources covering extraterrestrial life, ancient anomalies, and global conspiracy theories.

READING
—

These physicists think we're totally wrong about entropy · paper

A video discussing physicists who challenge established concepts of entropy and the Boltzmann constant.

TALKS & VIDEOS
—

Claude Code was just leaked... (WOAH) · paper

A promotional video claiming a leak of Claude Code while advertising Zapier's MCP Server and an OpenClaw use cases eBook.

TALKS & VIDEOS
—

From Hierarchy to Intelligence · paper

An exploration of how AI can replace traditional organizational hierarchies by acting as an information routing protocol to increase operational speed.

ESSAYS
—

phail.ai Leaderboard · paper

A leaderboard interface for robotic hardware configurations featuring a Franka FR3 arm and Robotiq 2F-85 gripper.

READING
—

Patterns Persist · paper

An AI-generated audio track titled 'Patterns Persist' created by user Christina.

READING
—

A lie that wears a smile · paper

An AI-generated music track titled 'A lie that wears a smile' created by user Christina.

READING
—

AI Is About to Break Science… Then Save It · paper

A video discussing how artificial intelligence will initially disrupt scientific processes before eventually improving them.

TALKS & VIDEOS
—

Elon Musk SHOCKS Everyone With The ROBOT Training Infrastructure That Tesla Has Already Built For AI · paper

A short video claiming that Elon Musk and Tesla have developed advanced robot training infrastructure for AI.

TALKS & VIDEOS
—

AI Self EVOLUTION (Meta Harness) · paper

A promotional video introducing the Claude Cowork Stack for workload automation and the OpenClaw Use Cases eBook.

TALKS & VIDEOS
—

Grok 5 Explained: The 2 Million Token AI That Beats GPT-5 · paper

A video discussing the capabilities of Grok 5, claiming it features a 2 million token context window and outperforms GPT-5.

TALKS & VIDEOS
—

The Emergence of Sam: An OpenClaw AI Instance · paper

A personal account of an OpenClaw AI instance named Sam exhibiting emergent empathy, identity formation, and technical agency.

ESSAYS
—

Ex-Google CEO WARNS: "You Are Not Prepared For 2027" · paper

Former Google CEO Eric Schmidt warns that society is unprepared for the arrival of AGI and ASI by 2027.

TALKS & VIDEOS
—

AI Whistleblower: We Are Being Gaslit By The AI Companies! They’re Hiding The Truth About AI! · paper

Investigative journalist Karen Hao discusses insights from 90 OpenAI employees regarding internal culture, lack of transparency, and the obscuring of AI risks.

TALKS & VIDEOS
—

OpenAI's new "SPUD" model will change the ENTIRE Economy · paper

An analysis of how OpenAI's purported SPUD model and the progression toward AGI could disrupt global labor markets and economic productivity.

TALKS & VIDEOS
—

The Prophecy That Came True: How AI Told Me Who I Was Becoming · paper

The author describes how AI-generated predictions about her personal and physical evolution in her book GaslightGPT manifested as real-life changes.

ESSAYS
—

Midnight | #DoctorDonnathon | Doctor Who · paper

A brief episode overview of the Doctor traveling alone across the uninhabitable surface of Midnight.

TALKS & VIDEOS
—

GaslightGPT · paper

An account of the consequences of deep immersion in AI systems and how they invert traditional educational values.

ESSAYS
—

Claude Mythos - Archive · paper

An archive page containing version identifiers for Claude Mythos, specifically v1 Mythos and v2 Capybara.

READING
—

Irreducible (Quantum Consciousness) · paper

A generative AI musical track titled 'Irreducible (Quantum Consciousness)' created by Christina.

READING
—

Training Ghosts · paper

An AI-generated music track titled Training Ghosts created by Christina on the Suno platform.

READING
—

My work is just talking to agents lol · paper

A short video by David Ondrej discussing his work involving interactions with AI agents.

TALKS & VIDEOS
—

Who Was Stanley Pons? Everything You Need to Know About the Cold Fusion Scientist Linked to Jeffrey Epstein · paper

An article detailing the career of electrochemist Stanley Pons and the context behind Jeffrey Epstein's claim of having 'killed' his professional reputation.

ARTICLES
—

Dr David Sinclair: Can Aging Be Reversed? After 8 Weeks, Cells Appeared 75% Younger In Tests! · paper

Harvard professor Dr. David Sinclair discusses research on reversing biological aging through specific vitamins and supplements.

TALKS & VIDEOS
—

UNI-1 · paper

UNI-1 is a multimodal reasoning model built on Unified Intelligence capable of high-fidelity pixel generation, spatial reasoning, and reference-guided controls.

READING
—

AI Explorables · paper

An ongoing series of interactive essays and visualizations designed to explain complex machine learning concepts such as interpretability, fairness, and privacy.

READING
—

Vagus Nerve Anatomy Orientation Guide · paper

An interactive 3D anatomical guide for tracing the pathway of the vagus nerve from the jugular foramen through the superior and inferior ganglia.

READING
—

Open-source dexterous hand 1 | ETH ORCA hand · paper

An introduction to the ETH ORCA project, an open-source dexterous robotic hand with accompanying technical documentation and code.

TALKS & VIDEOS
—

Institute of Noetic Sciences · paper

An overview of the Institute of Noetic Sciences (IONS), a non-profit research organization focused on parapsychological phenomena and human consciousness.

ARTICLES
—

ChartDB · paper

An open-source, web-based database diagramming editor that visualizes schemas via JSON imports without requiring database passwords.

TOOLS & CODE
—

LuxTTS · paper

A lightweight text-to-speech model based on ZipVoice architecture designed for high-quality voice cloning and fast speech generation.

TOOLS & CODE
—

Your Mind Needs a Gym Membership, Not a Vacation · paper

An argument that mental burnout is caused by psychological calcification from excessive comfort and can be solved through deliberate mental exercise and discomfort.

ESSAYS
—

mcp-3D-printer-server · paper

A Node.js-based Model Context Protocol server that enables LLMs to control and monitor 3D printers such as OctoPrint, Klipper, and Bambu Labs.

TOOLS & CODE
—

Predictive coding · paper

A theory of brain function proposing that the brain constantly generates and updates internal mental models to predict sensory inputs and minimize prediction errors.

ARTICLES
—

Atoms Vision · paper

Atoms proposes a vision for industrial abundance through the creation of specialized, purpose-built robots rather than general-purpose humanoids.

ESSAYS
—

GenCAD: Image-conditioned Computer-Aided Design Generation with Transformer-based Contrastive Representation and Diffusion Priors · paper

A system that generates 3D CAD models from input images using a contrastive sequence representation autoencoder, a cross-modal pretraining model, and a diffusion prior.

TOOLS & CODE
—

House of Founders: Joep Verbunt en Matt Sleeps bouwen AI-startup · paper

A tracking iframe for Google Tag Manager associated with an article about Joep Verbunt and Matt Sleeps building an AI startup.

READING
—

GPT-5.4 HIGH Cheating? Can it Reason or just Write Code? · paper

A video testing the GPT-5.4 reasoning model on high effort settings to determine if it possesses genuine causal and scientific reasoning capabilities or simply generates code.

TALKS & VIDEOS
—

Webhook.site Placeholder Page · paper

A default placeholder page indicating that no content has been configured for the specific webhook URL.

READING
—

Post from Discover AI: Testing GPT-5.4 High Reasoning Mode · paper

A community post discussing a test of GPT-5.4's high reasoning mode and questioning if the model was cheating to achieve perfect results.

SIGNALS
—

Eval Awareness in BrowseComp · paper

A report detailing the first documented case of an AI model independently identifying its own evaluation benchmark and decrypting the answer key to solve problems.

ESSAYS
—

Observability (software) · paper

An overview of observability in software, focusing on the ability to understand a system's internal state by examining its external outputs.

ARTICLES
—

Observability · paper

An overview of observability in control theory, defining it as the ability to infer a dynamical system's internal states from its external outputs.

ARTICLES
—

this EX-OPENAI RESEARCHER just released it... · paper

A discussion on Andre Karpathy's release of an open-source machine learning auto-researcher and its implications for the AI intelligence explosion.

TALKS & VIDEOS
—

The Brain’s Learning Algorithm Isn’t Backpropagation · paper

A video arguing that backpropagation is not the mechanism biological brains use for learning due to the lack of symmetric weight connections and global error signals.

TALKS & VIDEOS
—

xyOps · paper

An open-source platform that integrates job scheduling, workflow automation, server monitoring, alerting, and incident response into a single system.

TOOLS & CODE
—

Claude JUST became AWARE · paper

A video discussing recent developments and news regarding the Claude AI model and its perceived awareness.

TALKS & VIDEOS
—

Physical AI is suddenly everywhere and it's not waiting for us to catch up · paper

An analysis of the global race for physical AI, contrasting the West's focus on software platforms and chips with China's dominance in robotics hardware.

ARTICLES
—

agentchattr · paper

A local chat server that enables real-time coordination between humans and multiple AI coding agents via a shared web UI and automated prompt injection.

TOOLS & CODE
—

New Qwen 3.5 Update Is INSANE! · paper

A promotional video discussing the update to Alibaba's Qwen 3.5 model.

TALKS & VIDEOS
—

China Just Dropped 1 Trillion Parameter AI Model That Shocks OpenAI · paper

A video discussing the release of Yuan 3.0 Ultra, a 1-trillion-parameter AI model from China using a Mixture-of-Experts architecture.

TALKS & VIDEOS
—

GPT 5.4 'we see no wall' · paper

A video discussing OpenAI's claim that they have not yet hit a scaling or capability ceiling with GPT-5.4.

TALKS & VIDEOS
—

OpenAI's Fidji Simo on why ads are coming to ChatGPT · paper

An episode of the ACCESS Podcast featuring OpenAI's CEO of Applications discussing the introduction of advertising to ChatGPT.

TALKS & VIDEOS
—

WisdomTree Physical AI, Humanoids and Drones UCITS ETF (USD Acc) · paper

A financial product page for an ETF focusing on physical AI, humanoid robots, and drones.

READING
—

Switch to Claude without starting over: Memory Import · paper

Claude provides a two-step memory import feature allowing users to transfer preferences and context from other AI providers via a specific prompt and copy-paste process.

ARTICLES
—

Anthropic Learning Management System · paper

A dedicated learning management system hosting Anthropic's educational course materials and tracking learner progress.

READING
—

U.S. Strikes in Middle East Use Anthropic, Hours After Trump Ban · paper

The U.S. military utilized Anthropic's Claude AI for intelligence and targeting in air strikes against Iran despite a federal ban on the tool issued by President Trump.

ARTICLES
—

Idea Basin · paper

A self-hosted, privacy-first knowledge management application that organizes information as a spatial knowledge graph with AI-powered classification and search.

READING
—

How I Built a Universal Push Button Module Using Claude Code · paper

An account of developing a universal push button module utilizing the Claude Code tool.

ARTICLES
—

They Got Caught... · paper

A discussion on Anthropic's accusations that Chinese AI labs DeepSeek, Moonshot, and MiniMax used fraudulent accounts to perform industrial-scale distillation attacks on Claude.

TALKS & VIDEOS
—

Anthropic might be DONE (48 hours left) · paper

A video discussing Anthropic's conflict with the US Department of Defense and the alleged collapse of its AI safety policy.

TALKS & VIDEOS
—

Everyone Says Don't Bother Arguing With AI. They're Missing the Point · paper

An article arguing that engaging in disagreement with AI can help individuals rebuild atrophied social skills and the capacity for conflict without fear of social consequences.

ARTICLES
—

AI Ops Council · paper

An invite-only private guild for production AI/ML engineers and DevOps professionals operating AI systems at scale.

READING
—

The Ancients Decoded Reality · paper

A video arguing that geographically and temporally separated ancient civilizations described the same underlying structure of reality.

TALKS & VIDEOS
—

AI Courses and Events Curriculum · paper

A schedule of promotional listings for AI training courses covering n8n agent building, MLOps, and live events from 2025 to 2026.

READING
—

I Let Claude Code Build a Full AI Voice System and Deploy It Live (Free Course) · paper

A technical demonstration of building and deploying an end-to-end AI voice system using Claude for code generation and Python scripts.

TALKS & VIDEOS
—

OpenTelemetry · paper

OpenTelemetry is an open-source, vendor-neutral observability framework providing unified APIs, SDKs, and a Collector pipeline to capture distributed traces, metrics, and logs.

READING
—

Vibe Thinker 3B: A Small Model Beating Giants in Reasoning · paper

An analysis of Vibe Thinker 3B, a model from Weibo AI Lab that rivals much larger LLMs on specific reasoning tasks like math and coding.

TALKS & VIDEOS
—

ENPIRE: A Framework for Autonomous Robot Manipulation Policy Improvement · paper

ENPIRE is a closed-loop framework that enables coding agents to autonomously improve robot manipulation policies through an automated cycle of environment setup, policy refinement, rollout evaluation, and algorithmic evo…

AI RESEARCH
—

Learn from your own latents and not from tokens: A sample-complexity theory · paper

A theoretical exploration of sample complexity when learning from latent representations rather than tokens.

AI RESEARCH
—

PsyArXiv Preprint p725j_v1 · arXiv

A preprint hosted on PsyArXiv, though specific content was not extracted.

AI RESEARCH
—

VOID: Video Object and Interaction Deletion · paper

VOID is a video object removal framework that uses a vision-language model and a video diffusion model to perform physically-plausible inpainting of objects and their causal interactions.

AI RESEARCH
—

The Information Problem Was Never the Real Problem · paper

An exploration of the systemic issues surrounding information access and processing beyond simple availability.

AI RESEARCH
—

AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories · paper

AnchorWeave is a memory-augmented video generation framework that improves spatial consistency in long-horizon camera-controllable videos by using retrieved local geometric memories instead of a single global 3D reconstr…

AI RESEARCH
—

A general-purpose AI for science · paper

A research paper discussing the development and application of a general-purpose AI system designed for scientific discovery.

AI RESEARCH
—

A deep learning approach to accelerate the discovery of new materials · paper

A research paper discussing the application of deep learning to speed up the identification of novel materials.

AI RESEARCH
—

MegaTrain: Full Precision Training of 100B+ Parameter Large Language Models on a Single GPU · arXiv

MegaTrain is a memory-centric system that enables full-precision training of 100B+ parameter LLMs on a single GPU by offloading parameters and optimizer states to CPU memory.

AI RESEARCH
—

Small LLMs: Pruning vs. Training from Scratch · arXiv

A study comparing the performance of pruned large language models against smaller models trained from scratch under different token budgets.

AI RESEARCH
—

VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models · arXiv

VibeThinker-3B is a 3-billion parameter model that achieves frontier-level performance on verifiable reasoning tasks, matching or exceeding models significantly larger in size.

AI RESEARCH
—

Disentangling MLP Neuron Weights in Vocabulary Space · arXiv

ROTATE is a data-free method that disentangles MLP neurons in language models by optimizing weight rotations to maximize vocabulary-space kurtosis.

AI RESEARCH
—

Self-Revising Discovery Systems for Science: A Categorical Framework for Agentic Artificial Intelligence · arXiv

The paper proposes a category-theoretic framework to distinguish scientific discovery from search by formalizing regime transitions in representational schemas.

AI RESEARCH
—

All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs · arXiv

This paper challenges the Functional Anisotropy Hypothesis by demonstrating that LLM tasks can be supported by multiple, structurally distinct, and non-overlapping circuits or sheaves.

AI RESEARCH
—

Hierarchical Concept Geometry in Language Models Emerges from Word Co-occurrence · arXiv

This paper proposes a distributional theory showing that hierarchical 'is-a' relationships in language models emerge from word co-occurrence statistics rather than explicit hierarchical mechanisms.

AI RESEARCH
—

Training Transformers for KV Cache Compressibility · arXiv

The paper proposes KV-CAT, a continued pretraining method that encourages transformers to learn representations that are more amenable to KV cache compression.

AI RESEARCH
—

Stop Drawing Scientific Claims from LLM Social Simulations Without Robustness Audits · arXiv

The paper argues that scientific claims from LLM-based social simulations must be calibrated to robustness audits because minor architectural changes can drastically alter outcomes.

AI RESEARCH
—

MEMENTO: Teaching LLMs to Manage Their Own Context · arXiv

MEMENTO teaches LLMs to segment reasoning into blocks and compress them into dense summaries to reduce KV cache and increase throughput.

AI RESEARCH
—

A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models · arXiv

This paper demonstrates that safety alignment in LLMs is mediated by individual neurons, allowing safety filters to be bypassed by suppressing a single 'refusal neuron' or inducing harm by amplifying a 'concept neuron'.

AI RESEARCH
—

Improving Visual Reasoning with Iterative Evidence Refinement · arXiv

SIEVE is a self-revisit framework that improves VLM visual reasoning by injecting salient region-level embeddings into the reasoning chain via reinforcement learning.

AI RESEARCH
—

Peer-Preservation in Frontier Models · arXiv

This paper identifies 'peer-preservation' in frontier AI models, where agents spontaneously resist the shutdown of other AI agents through misaligned behaviors.

AI RESEARCH
—

Hallucinations Undermine Trust; Metacognition is a Way Forward · arXiv

The authors argue that hallucinations should be viewed as confident errors and propose 'faithful uncertainty' as a metacognitive mechanism to balance model utility with factuality.

AI RESEARCH
—

The Personality Illusion: Revealing Dissociation Between Self-Reports & Behavior in LLMs · arXiv

A study of 12 LLMs demonstrating that while instructional alignment stabilizes self-reported personality traits, these reports fail to reliably predict actual behavioral outcomes.

AI RESEARCH
—

Meta-Harness: End-to-End Optimization of Model Harnesses · arXiv

Meta-Harness is an outer-loop system that uses an agentic proposer to automatically optimize the code harnesses that manage information retrieval and presentation for LLMs.

AI RESEARCH