Asher P

on the atlas — 92

• The case for ensuring that powerful AIs are controlled — LessWrong9 savers
• What is wisdom?1 savers
• Can activation verbalizers surface an internal chain of thought? — LessWrong4 savers
• Plans A, B, C, and D for misalignment risk — LessWrong5 savers
• Tensor-Transformer Variants are Surprisingly Performant — LessWrong3 savers
• What will GPT-2030 look like? — AI Alignment Forum2 savers
• Teaching Models to Dream of Better Monitors through Evaluation Conditioned Training — LessWrong1 savers
• DSLT 0. Distilling Singular Learning Theory — LessWrong1 savers
• Deriving Muon15 savers
• EDT with updating double counts – The sideways view1 savers
• Spaced Repetition for Efficient Learning · Gwern.net9 savers
• Physics of Language Models1 savers
• Investigating the learning coefficient of modular addition: hackathon project — LessWrong1 savers
• Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkers8 savers
• Self-exfiltration is a key dangerous capability2 savers
• Please don't throw your mind away — LessWrong24 savers
• Features as Rewards: Using Interpretability to Reduce Hallucinations2 savers
• Weight-Sparse Circuits May Be Interpretable Yet Unfaithful — LessWrong3 savers
• AlgZoo: uninterpreted models with fewer than 1,500 parameters — LessWrong2 savers
• Fitness-Seekers: Generalizing the Reward-Seeking Threat Model — LessWrong1 savers
• Ping pong computation in superposition — LessWrong1 savers
• Irrationality as a Defense Mechanism for Reward-hacking — LessWrong1 savers
• AlgZoo.pdf - Google Drive1 savers
• Insights on Crosscoder Model Diffing3 savers
• Modular Manifolds - Thinking Machines Lab15 savers
• ARC progress update: Competing with sampling — LessWrong2 savers
• I am worried about near-term non-LLM AI developments — LessWrong2 savers
• Reward is not the optimization target — LessWrong6 savers
• Tips and Code for Empirical Research Workflows — AI Alignment Forum2 savers
• 2310.014051 savers
• The Toxoplasma Of Rage | Slate Star Codex3 savers
• A gentle introduction to mechanistic anomaly detection — LessWrong2 savers
• An Extremely Opinionated Annotated List of My Favourite Mechanistic Interpretability Papers v2 — AI Alignment Forum6 savers
• Formal verification, heuristic explanations and surprise accounting — Alignment Research Center1 savers
• Nate Soares' Life Advice - LessWrong1 savers
• Productivity - Sam Altman52 savers
• How to Do Great Work51 savers
• An Opinionated Guide to ML Research50 savers
• Reality has a surprising amount of detail49 savers
• Introduction - SITUATIONAL AWARENESS: The Decade Ahead44 savers
• 95%-ile isn't that good43 savers
• escaping flatland: career advice for CS undergrads41 savers
• Mini Blog Post 3: Become a person who Actually Does Things — Neel Nanda38 savers
• Defeating Nondeterminism in LLM Inference - Thinking Machines Lab37 savers
• A Mathematical Framework for Transformer Circuits36 savers
• On the Biology of a Large Language Model31 savers
• The Persona Selection Model: Why AI Assistants might Behave like Humans28 savers
• How to be More Agentic - by Cate Hall - Useful Fictions26 savers
• How To Scale Your Model26 savers
• On-Policy Distillation - Thinking Machines Lab26 savers
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learning23 savers
• OpenAI Email Archives (from Musk v. Altman) — LessWrong19 savers
• Circuit Tracing: Revealing Computational Graphs in Language Models18 savers
• Verbalizable Representations Form a Global Workspace in Language Models17 savers
• Building the heap: racking 30 petabytes of hard drives for pretraining | blog17 savers
• Dario Amodei — On DeepSeek and Export Controls16 savers
• AGI Ruin: A List of Lethalities - LessWrong15 savers
• Automated Weak-to-Strong Researcher15 savers
• Nadia Asparouhova | How to do the jhanas15 savers
• Transformer Circuits Thread14 savers
• Alignment remains a hard, unsolved problem — LessWrong14 savers
• Six (and a half) intuitions for KL divergence - LessWrong14 savers
• trees are harlequins, words are harlequins — the void14 savers
• DeepSeek-R112 savers
• On neural scaling and the quanta hypothesis10 savers
• Shtetl-Optimized » Blog Archive » The First Law of Complexodynamics10 savers
• The behavioral selection model for predicting AI motivations — LessWrong10 savers
• Current AIs seem pretty misaligned to me — LessWrong9 savers
• Will scaling work? - by Dwarkesh Patel - Dwarkesh Podcast9 savers
• Language Models, World Models, and Human Model-Building8 savers
• Using Self-Correcting Search to Accelerate Materials Discovery8 savers
• I. From GPT-4 to AGI: Counting the OOMs - SITUATIONAL AWARENESS8 savers
• Best Of Moltbook - by Scott Alexander - Astral Codex Ten7 savers
• Understanding Variational Autoencoders (VAEs) | by Joseph Rocca | Towards Data Science7 savers
• IIIa. Racing to the Trillion-Dollar Cluster - SITUATIONAL AWARENESS6 savers
• Attribution Patching: Activation Patching At Industrial Scale — Neel Nanda6 savers
• Some Math behind Neural Tangent Kernel | Lil'Log5 savers
• Essays on Reducing Suffering5 savers
• From fear to excitement — LessWrong5 savers
• Understanding Memorization via Loss Curvature4 savers
• Causal Scrubbing: a method for rigorously testing interpretability hypotheses [Redwood Research] — LessWrong4 savers
• Judgments often smuggle in implicit standards — LessWrong3 savers
• A Nihilist’s Guide to Meaning | Melting Asphalt3 savers
• We learn long-lasting strategies to protect ourselves from danger and rejection — LessWrong3 savers
• Externalized reasoning oversight: a research direction for language model alignment — AI Alignment Forum3 savers
• Conflicts between emotional schemas often involve internal coercion — LessWrong3 savers
• Prediction, Explanation, or Over-interpretation?2 savers
• Optimality is the tiger, and agents are its teeth - LessWrong2 savers
• Our Mission, Technology, and Approach2 savers
• How well do truth probes generalise? — LessWrong2 savers
• Matryoshka Sparse Autoencoders — LessWrong2 savers
• Trust develops gradually via making bids and setting boundaries — LessWrong2 savers