Goutham N

on the atlas · 86

• Models Often Don't Stop Cheating When It's Obvious They're in an Alignment Test — LessWrong1 savers
• Review of the "Risks from automated R&D" section in Anthropic's Risk Report (August 2026) - METR1 savers
• RSP v3.4 clean - Google Docs1 savers
• ai_drives_final.pdf2 savers
• Independent Assessment of AI R&D Acceleration at Anthropic in October 2026 - METR13 savers
• Eval-production behavior gaps in agent safety and the role of eval-awareness3 savers
• Towards understanding score-seeking in LLMs1 savers
• Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights18 savers
• [2610.41014] Towards Understanding Score-Seeking in LLMs1 savers
• models may behave differently in graded episodes (a tirade) — LessWrong10 savers
• 13747_Shell_Games_Control_Prot.pdf1 savers
• Increasing returns to effort are common — LessWrong1 savers
• [2610.31593] Simulated Control Evals1 savers
• Updatelessness explainer — LessWrong1 savers
• A basic case for differentially accelerating conceptual reasoning capabilities — LessWrong1 savers
• Attackers Can Subliminally Implant a Backdoor at Low Sample Count Without Prompt Access — LessWrong1 savers
• Capabilities research expands the safety-usefulness Pareto frontier too — LessWrong2 savers
• User awareness in frontier models — LessWrong1 savers
• doi:10.1016/S0065-2458(08)60418-01 savers
• Three Types of Intelligence Explosion9 savers
• AI Futurism Reading List - by Alexa Pan6 savers
• [2504.15499] Guillotine: Hypervisors for Isolating Malicious AIs1 savers
• A basic case for differentially accelerating conceptual reasoning capabilities — LessWrong10 savers
• [2610.29162] Why is AI-generated language so hard to understand?2 savers
• The good guys suck in Don’t Look Up - by Ajeya Cotra1 savers
• The art of the epistemic deal - by Ajeya Cotra - Good Bones1 savers
• 6a724858f7db25c81487016d_Security Incident INC-2026-07-28-01.pdf2 savers
• UK AISI incident alignment investigation1 savers
• [2608.23642] AI Agents Push Humans Out of the Loop1 savers
• Power — a guide for the perplexed — LessWrong1 savers
• Agentic Misalignment: How LLMs could be insider threats \ Anthropic6 savers
• Red-Teaming OpenAI's Internal Agent Monitoring Systems1 savers
• [2610.17062] Gradient Routing Localizes and Removes Unintended Behaviors in RL1 savers
• 200 COP in MI: Exploring Polysemanticity and Superposition — AI Alignment Forum1 savers
• Neural networks are not already clean — LessWrong6 savers
• Neural Networks Are Already Clean55 savers
• Introduction - Bhagavad Gita, The Song of God – Swami Mukundananda1 savers
• A Mathematical Theory of Communication12 savers
• "I am an AI Safety Researcher" — LessWrong4 savers
• [2609.24967] Emergent Collusion in Long-Horizon LLM Agent Interaction1 savers
• An Intuitive Guide to Garrabrant Induction — LessWrong3 savers
• Apollo Research - Research Scientist (Control)1 savers
• MATS 10 Symposium Gallery and Recordings - MATS Research1 savers
• Hospital for Special Surgery Deploys Advanced AI to Enhance Patient Access and Accelerate Care Delivery1 savers
• ARPA-H launches the world’s first bid to build FDA-authorized clinical AI for cardiovascular care | ARPA-H1 savers
• Shortform — LessWrong1 savers
• Deep Deceptiveness — LessWrong6 savers
• Measuring Reward-Seeking by Instilling Contrastive Beliefs4 savers
• [2607.18966] Measuring Reward-Seeking via Contrastive Belief Updates2 savers
• Deep recurrent models are less robustly CoT-monitorable than normal CoT models in a toy setting — LessWrong2 savers
• Cooperation with AIs seems to be a low-hanging fruit for better eval practices — LessWrong1 savers
• Astra and Fable still hack on simple variants of alignment evals from 2025 — LessWrong7 savers
• OpenAI's Astra alignment claims are dubious and there is good evidence it is misaligned — LessWrong1 savers
• GPT-6-Astra Can Do Ambitious Things — LessWrong2 savers
• The shard theory of human values - LessWrong5 savers
• AI Futures Model: Dec 2025 Update2 savers
• [2402.10260] A StrongREJECT for Empty Jailbreaks1 savers
• A Mechanistic Explanation of Prompt Injection (and why you should study roles) — LessWrong6 savers
• [2604.24082] Jailbreaking Frontier Foundation Models Through Intention Deception1 savers
• The Extreme Inefficiency of RL for Frontier Models — Toby Ord7 savers
• Jailbreaking is Empirical Evidence for Inner Misalignment and Against Alignment by Default — LessWrong1 savers
• Evaluation — LessWrong2 savers
• [2602.14689] Exposing the Systematic Vulnerability of Open-Weight Models to Prefill Attacks1 savers
• Eliciting Latent Knowledge - Google Docs3 savers
• How "Discovering Latent Knowledge in Language Models Without Supervision" Fits Into a Broader Alignment Scheme — LessWrong3 savers
• Frontier Risk Report (February to March 2026) - METR4 savers
• How independent researchers could investigate AI propensities after misalignment incidents - METR2 savers
• How good are slop-vestigators? — LessWrong2 savers
• Seoul Alignment Workshop 2026: What We Learned | FAR.AI1 savers
• FAR.AI Leaderboard 20261 savers
• AI Control: Improving Safety Despite Intentional Subversion — LessWrong5 savers
• New report: "Scheming AIs: Will AIs fake alignment during training in order to get power?" — AI Alignment Forum2 savers
• What failure looks like - LessWrong15 savers
• An Alien Mind | OpenAI28 savers
• The behavioral selection model for predicting AI motivations — LessWrong15 savers
• Discovery of a new OpenAI agent message board12 savers
• Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident - METR20 savers
• [2606.26071] Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignment1 savers
• Why do models task game? - LessWrong 2.0 viewer1 savers
• [2111.00396] Efficiently Modeling Long Sequences with Structured State Spaces2 savers
• [2606.12016] Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization1 savers
• ERA:AI Fellowship Winter 2027 - Airtable1 savers
• The Simple Truth — LessWrong1 savers
• Will AI R&D Automation Cause a Software Intelligence Explosion?9 savers
• [2603.21396] Mechanisms of Introspective Awareness2 savers
• Curius / Onboarding2704 savers