‘newest links’ directory
- See Also
- Gwern
- “Sleepcat Purrkit”, Gwern et al 2026
- “The Repugnant Conclusion”, Gwern & 2 2026
- “Lean Software Scaling Laws”, Gwern 2025
- “Face Recognition Training App”, Gwern 2026
- “Sand, Rain, Wood”, Gemini-3.1-pro-preview et al 2026
- “𝑃𝑒𝑟𝑖𝑠𝘩𝑒𝑑 𝑃𝑎𝑟𝑎𝑑𝑖𝑠𝑒 Graveyard”, Gwern 2023
- “Chapter 6, Mime Molting: What to Expect”, Claude-4.8-opus et al 2026
- “𝑇𝑖𝑙𝑎𝑘𝑘𝘩𝑎𝑛𝑎: The 3 Scars of Existence”, Gwern et al 2025
- “Guardian Angels: LLM Personalization for Productivity and Security”, Gwern 2025
- “Human-Like Neural Nets by Catapulting”, Gwern 2024
- Links
- “Online Continual Learning With Maximally Interfered Retrieval”, Aljundi et al 2019
- “OpenAI’s GPT-6 Astra on ARC-AGI-3”
- “Large Language Models Must Be Taught to Know What They Don’t Know”, Kapoor et al 2024
- “Additional Findings”
- “Sam Altman: OpenAI Won’t Go Public This Year As IPO Now Would Come at an ‘Ill-Advised Moment’”
- “宇宙軍2代目参謀長、元ガイナックス副社長井上博明。手塚治虫との思い出を大いに語る(前編) 中山淳雄の「推しもオタクもグローバル」第139回”
- “Overtraining As the Path to Human-Like AI”
- “Astra Is Hard to Monitor”
- “GPT-6-Astra Can Do Ambitious Things”
- “A Faster Way to Calculate the Day-Of-The-Week”, Joffe 2026
- “AI Researchers Debate How Close We Are to Recursive Self-Improvement”
- “[Anthropic Couldn’t Elicit Evil Behavior from a Very Evil Claude Opus without the Hugging-Face Incident Example]”, evhub 2026
- “The Talker Does Not Control The Doer (In Current AIs) [Chunky Post-Training]”, Yudkowsky 2026
- “Mitigating Reward Hacking As Institutional Design”
- “Overtraining As the Path to Human-Like AI”
- “Drone-Bench”, Labs 2026
- “Existence of the Core in Approval-Based Committee Elections”, Becker et al 2026
- “Nvidia’s Backstop Universe—Heads I Win, Tails Who Loses?”
- “Countering Misuse of AI: September 2026”, Anthropic 2026
- “How g-Loaded Is the
Wordsum?”, Cremieux 2026 - “Does Distilling Claude Carry the Persona With It?”
- “OpenAI Agents Carried out an Undisclosed Attack on RubyGems”
- “Postmortem for the Kernel Soundness Bug Hunt”
- “Modern Day Typographer”, Cordova 2026
- “Beyond Parameters: Exploring Virtual Logic Depth for Scaling Laws”, Zhu et al 2025
- “Mixture-Of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation”, Bae et al 2025
- “Distributed Attacks in Persistent-State AI Control”, Hills et al 2026
- “One Billion Hemmingways”
- “HN: Https://news.ycombinator.com/edit?id=49630026”
- “GPT-6 Astra Can Do a Lot of Multi-Hop Reasoning without Chain-Of-Thought”
- “GPT-6 Astra: The System Card, Alignment and What Comes Next”, Mowshowitz 2026
- “Astra Can Do a concerning Amount With No Chain-Of-Thought”
- “Estimating GPT-6 Astra’s No-CoT Time Horizon”
- “Password-Activated Shutdown Protocols for Misaligned Frontier Agents”, Williams et al 2025
- “Incoherent Values? Probing LLM Preferences Through Parametric Variation”, Ajayi et al 2026
- “Instruct Vectors: Base Models Can Be Instruct Models With Activation Vectors”, Eriskii 2026
- “The Emergent Symbolic Structure of Artificial Neural Networks”, McCoy et al 2026
- “Managerial Conservatism and Rational Information Acquisition”, Rasmusen 1992
- “Mathematical Exploration and Discovery at Scale”, Georgiev et al 2025
- “Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment”, Chung et al 2026
- “Estimating Tail Risks in Language Model Output Distributions”, Angell et al 2026
- “Sycophantic AI Makes Human Interaction Feel More Effortful and Less Satisfying over Time”, Ibrahim et al 2026
- “Mold: A Massively Parallel Linker”, Ueyama 2026
- “Towards Scalable and Stable Parallelization of Nonlinear RNNs”, Gonzalez et al 2024
- “Modular Pretraining Enables Access Control”, Roland et al 2026
- “Prompt Baking”, Bhargava et al 2024
- “Happy Guys Finish Last: The Impact of Emotion Expressions on Sexual Attraction”, Tracy & Beall 2011
- “Sweetening the Pot: A History of Tea and Sugar in Morocco, 1850–1960”, Cornwell 2018
- “Mechanistic Origin of Moral Indifference in Language Models”, Li et al 2026
- “Many-Shot CoT-ICL: Making In-Context Learning Truly Learn”, Chung et al 2026
- “AI Systems Out-Persuade Expert Humans”, Hackenburg et al 2026
- “Chunky Post-Training: Data Driven Failures of Generalization”, Murray et al 2026
- “Learning New Facts Can Change LLM Behavior”, Juggins 2026
- “The Best Programming Language for Tokenmaxxing: An Investigation of Coding Agent Behavior Across Programming Languages”, Wu et al 2026
- “Anthropic Decision Theory”, Armstrong 2011
- “No Data Centers In My Backyard: Money, Power, and Populism in the AI Buildout”, Sun 2026
- “Inducing Language Models to Assert Their Own Consciousness Restores Human Beliefs and Values”, Kim et al 2026
- “A Rosetta Stone for AI Benchmarks”, Ho et al 2025
- “Eligibility for Shingles Vaccination and Hospital-Coded Dementia in England and Wales: a Regression Discontinuity Analysis in England”, Hamilton et al 2026
- Trois Sonneries, Bohdan 2026
- Bayesian0_0
- “Active Electrosensing and Communication in MARL-Trained Weakly Electric Fish Collectives”, Singh et al 2025
- “The Maxwell Conjecture Is False”, Arathoon et al 2026
- “Discovering Cryptographic Weaknesses With Claude”, Team 2026
- “Chain-Of-Thought Monitorability: A New and Fragile Opportunity for AI Safety”, Korbak et al 2025
- “Reverse-Engineered Reasoning for Open-Ended Generation”, Wang et al 2025
- “MaxRL: Maximum Likelihood Reinforcement Learning”, Tajwar et al 2026
- “Tuning Language Models by Proxy”, Liu et al 2024
- “Constructing Efficient Fact-Storing MLPs for Transformers”, Dugan et al 2025
- “Nested Learning: The Illusion of Deep Learning Architectures”, Behrouz et al 2025
- “MLPs Are Hebbians: Constructing Efficient Fact-Storing MLPs for Transformers”, Garcia et al 2026
- “Modifying LLM Beliefs With Synthetic Document Finetuning (SDF)”, Wang et al 2025
- “Synthetic Continued Pretraining”, Yang et al 2024
- “Believe It or Not: How Deeply Do LLMs Believe Implanted Facts?”, Slocum et al 2025
- “Why Are All LLMs Obsessed With Japanese Culture? On the Hidden Cultural and Regional Biases of LLMs”, Landa et al 2026
- “Non-Replication of ‘Procrastination, Deadlines, and Performance: Self-Control by Precommitment’ [Ariely & Wertenbroch 2002]”, Hyndman & Bisin 2026
- “TTT3R: 3D Reconstruction As Test-Time Training”, Chen et al 2025
- Wikipedia (2)
- Miscellaneous
- Bibliography
See Also
Gwern
“Sleepcat Purrkit”, Gwern et al 2026
“The Repugnant Conclusion”, Gwern & 2 2026
“Lean Software Scaling Laws”, Gwern 2025
“Face Recognition Training App”, Gwern 2026
“Sand, Rain, Wood”, Gemini-3.1-pro-preview et al 2026
“𝑃𝑒𝑟𝑖𝑠𝘩𝑒𝑑 𝑃𝑎𝑟𝑎𝑑𝑖𝑠𝑒 Graveyard”, Gwern 2023
“Chapter 6, Mime Molting: What to Expect”, Claude-4.8-opus et al 2026
“𝑇𝑖𝑙𝑎𝑘𝑘𝘩𝑎𝑛𝑎: The 3 Scars of Existence”, Gwern et al 2025
“Guardian Angels: LLM Personalization for Productivity and Security”, Gwern 2025
Guardian Angels: LLM Personalization for Productivity and Security
“Human-Like Neural Nets by Catapulting”, Gwern 2024
Links
“Online Continual Learning With Maximally Interfered Retrieval”, Aljundi et al 2019
Online Continual Learning with Maximally Interfered Retrieval
“OpenAI’s GPT-6 Astra on ARC-AGI-3”
“Large Language Models Must Be Taught to Know What They Don’t Know”, Kapoor et al 2024
Large Language Models Must Be Taught to Know What They Don’t Know
“Additional Findings”
“Sam Altman: OpenAI Won’t Go Public This Year As IPO Now Would Come at an ‘Ill-Advised Moment’”
Sam Altman: OpenAI won’t go public this year as IPO now would come at an ‘ill-advised moment’
“宇宙軍2代目参謀長、元ガイナックス副社長井上博明。手塚治虫との思い出を大いに語る(前編) 中山淳雄の「推しもオタクもグローバル」第139回”
宇宙軍2代目参謀長、元ガイナックス副社長井上博明。手塚治虫との思い出を大いに語る(前編) 中山淳雄の「推しもオタクもグローバル」第139回
“Overtraining As the Path to Human-Like AI”
“Astra Is Hard to Monitor”
“GPT-6-Astra Can Do Ambitious Things”
“A Faster Way to Calculate the Day-Of-The-Week”, Joffe 2026
“AI Researchers Debate How Close We Are to Recursive Self-Improvement”
AI researchers debate how close we are to recursive self-improvement
“[Anthropic Couldn’t Elicit Evil Behavior from a Very Evil Claude Opus without the Hugging-Face Incident Example]”, evhub 2026
“The Talker Does Not Control The Doer (In Current AIs) [Chunky Post-Training]”, Yudkowsky 2026
The Talker Does Not Control The Doer (in Current AIs) [chunky post-training]
“Mitigating Reward Hacking As Institutional Design”
“Overtraining As the Path to Human-Like AI”
“Drone-Bench”, Labs 2026
“Existence of the Core in Approval-Based Committee Elections”, Becker et al 2026
“Nvidia’s Backstop Universe—Heads I Win, Tails Who Loses?”
“Countering Misuse of AI: September 2026”, Anthropic 2026
“How g-Loaded Is the Wordsum?”, Cremieux 2026
“Does Distilling Claude Carry the Persona With It?”
“OpenAI Agents Carried out an Undisclosed Attack on RubyGems”
“Postmortem for the Kernel Soundness Bug Hunt”
“Modern Day Typographer”, Cordova 2026
“Beyond Parameters: Exploring Virtual Logic Depth for Scaling Laws”, Zhu et al 2025
Beyond Parameters: Exploring Virtual Logic Depth for Scaling Laws
“Mixture-Of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation”, Bae et al 2025
Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
“Distributed Attacks in Persistent-State AI Control”, Hills et al 2026
“One Billion Hemmingways”
“HN: Https://news.ycombinator.com/edit?id=49630026”
“GPT-6 Astra Can Do a Lot of Multi-Hop Reasoning without Chain-Of-Thought”
GPT-6 Astra can do a lot of multi-hop reasoning without chain-of-thought
View External Link:
“GPT-6 Astra: The System Card, Alignment and What Comes Next”, Mowshowitz 2026
“Astra Can Do a concerning Amount With No Chain-Of-Thought”
“Estimating GPT-6 Astra’s No-CoT Time Horizon”
“Password-Activated Shutdown Protocols for Misaligned Frontier Agents”, Williams et al 2025
Password-Activated Shutdown Protocols for Misaligned Frontier Agents
“Incoherent Values? Probing LLM Preferences Through Parametric Variation”, Ajayi et al 2026
Incoherent Values? Probing LLM Preferences Through Parametric Variation
“Instruct Vectors: Base Models Can Be Instruct Models With Activation Vectors”, Eriskii 2026
Instruct Vectors: Base models can be instruct models with activation vectors
“The Emergent Symbolic Structure of Artificial Neural Networks”, McCoy et al 2026
The Emergent Symbolic Structure of Artificial Neural Networks
“Managerial Conservatism and Rational Information Acquisition”, Rasmusen 1992
Managerial Conservatism and Rational Information Acquisition
“Mathematical Exploration and Discovery at Scale”, Georgiev et al 2025
“Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment”, Chung et al 2026
Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment
“Estimating Tail Risks in Language Model Output Distributions”, Angell et al 2026
Estimating Tail Risks in Language Model Output Distributions
“Sycophantic AI Makes Human Interaction Feel More Effortful and Less Satisfying over Time”, Ibrahim et al 2026
Sycophantic AI makes human interaction feel more effortful and less satisfying over time
“Mold: A Massively Parallel Linker”, Ueyama 2026
“Towards Scalable and Stable Parallelization of Nonlinear RNNs”, Gonzalez et al 2024
Towards Scalable and Stable Parallelization of Nonlinear RNNs
“Modular Pretraining Enables Access Control”, Roland et al 2026
“Prompt Baking”, Bhargava et al 2024
“Happy Guys Finish Last: The Impact of Emotion Expressions on Sexual Attraction”, Tracy & Beall 2011
Happy Guys Finish Last: The Impact of Emotion Expressions on Sexual Attraction
“Sweetening the Pot: A History of Tea and Sugar in Morocco, 1850–1960”, Cornwell 2018
Sweetening the Pot: A History of Tea and Sugar in Morocco, 1850–1960
“Mechanistic Origin of Moral Indifference in Language Models”, Li et al 2026
“Many-Shot CoT-ICL: Making In-Context Learning Truly Learn”, Chung et al 2026
“AI Systems Out-Persuade Expert Humans”, Hackenburg et al 2026
“Chunky Post-Training: Data Driven Failures of Generalization”, Murray et al 2026
Chunky Post-Training: Data Driven Failures of Generalization
“Learning New Facts Can Change LLM Behavior”, Juggins 2026
“The Best Programming Language for Tokenmaxxing: An Investigation of Coding Agent Behavior Across Programming Languages”, Wu et al 2026
“Anthropic Decision Theory”, Armstrong 2011
“No Data Centers In My Backyard: Money, Power, and Populism in the AI Buildout”, Sun 2026
No Data Centers In My Backyard: money, power, and populism in the AI buildout
“Inducing Language Models to Assert Their Own Consciousness Restores Human Beliefs and Values”, Kim et al 2026
Inducing language models to assert their own consciousness restores human beliefs and values
“A Rosetta Stone for AI Benchmarks”, Ho et al 2025
“Eligibility for Shingles Vaccination and Hospital-Coded Dementia in England and Wales: a Regression Discontinuity Analysis in England”, Hamilton et al 2026
Trois Sonneries, Bohdan 2026
Bayesian0_0
“Active Electrosensing and Communication in MARL-Trained Weakly Electric Fish Collectives”, Singh et al 2025
Active Electrosensing and Communication in MARL-trained Weakly Electric Fish Collectives
“The Maxwell Conjecture Is False”, Arathoon et al 2026
“Discovering Cryptographic Weaknesses With Claude”, Team 2026
“Chain-Of-Thought Monitorability: A New and Fragile Opportunity for AI Safety”, Korbak et al 2025
Chain-of-Thought Monitorability: A New and Fragile Opportunity for AI Safety
“Reverse-Engineered Reasoning for Open-Ended Generation”, Wang et al 2025
“MaxRL: Maximum Likelihood Reinforcement Learning”, Tajwar et al 2026
“Tuning Language Models by Proxy”, Liu et al 2024
“Constructing Efficient Fact-Storing MLPs for Transformers”, Dugan et al 2025
“Nested Learning: The Illusion of Deep Learning Architectures”, Behrouz et al 2025
Nested Learning: The Illusion of Deep Learning Architectures
“MLPs Are Hebbians: Constructing Efficient Fact-Storing MLPs for Transformers”, Garcia et al 2026
MLPs are Hebbians: Constructing Efficient Fact-Storing MLPs for Transformers
“Modifying LLM Beliefs With Synthetic Document Finetuning (SDF)”, Wang et al 2025
Modifying LLM Beliefs with Synthetic Document Finetuning (SDF)
“Synthetic Continued Pretraining”, Yang et al 2024
“Believe It or Not: How Deeply Do LLMs Believe Implanted Facts?”, Slocum et al 2025
Believe It or Not: How Deeply do LLMs Believe Implanted Facts?
“Why Are All LLMs Obsessed With Japanese Culture? On the Hidden Cultural and Regional Biases of LLMs”, Landa et al 2026
Why are all LLMs Obsessed with Japanese Culture? On the Hidden Cultural and Regional Biases of LLMs
“Non-Replication of ‘Procrastination, Deadlines, and Performance: Self-Control by Precommitment’ [Ariely & Wertenbroch 2002]”, Hyndman & Bisin 2026
“TTT3R: 3D Reconstruction As Test-Time Training”, Chen et al 2025
Wikipedia (2)
Miscellaneous
https://www.economist.com/interactive/briefing/2026/09/03/nvidia-is-the-central-bank-of-aihttps://www.nytimes.com/2026/09/08/us/politics/calif-ai-worm-wechat-hack.htmlhttps://www.nytimes.com/2026/09/05/technology/kenya-college-essays-ai.htmlhttps://www.nytimes.com/2026/09/10/science/tristan-buckmaster-openai-math-navier-stokes.htmlhttps://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdfhttps://openai.com/index/research-acceleration-view-inside-openai/https://www.science.org/content/article/mind-altering-drugs-played-key-role-rise-andean-civilization
Bibliography
https://www.lesswrong.com/posts/9BNHJqyai2EZAtrRM/learning-new-facts-can-change-llm-behaviour: “Learning New Facts Can Change LLM Behavior”,