<
Home
Favorites
Strategy as a Wicked Problem (Camillus, 2008)
Why You Can’t Just Do Things - Octopusyarn
Strategic Taste
How to win a best paper award (or, an opinionated take on how to do important research)
How I've run major projects | benkuhn.net
Everything I've read
all
research
policy/fieldbuilding
self-improvement
culture
[p/f]
What We Learned from Briefing 140+ Lawmakers on the Threat from AI — LessWrong
[s-i]
The Best PR Advice You’ve Never Heard - from Facebook’s Head of Tech Communications
[p/f]
MIRI 2024 Communications Strategy - Machine Intelligence Research Institute
[c]
We Must Remember That Our World Contains Hell — LessWrong
[r]
RL creates split personas — LessWrong
[r]
Q2.5 2026 Timelines Update: Uplift and Revenue
[r]
A Spillway for Agent Coordination — LessWrong
[r]
Arguments for P — LessWrong
[r]
Why do models task game? — LessWrong
[r]
User awareness in frontier models | Transluce AI
[r]
This A.I. Just Created Viruses Not Found in Nature - The New York Times
[p/f]
How to pace the US frontier
[p/f]
The next chapter of our AI momentum
[r]
Returning to ARC — LessWrong
[r]
AI energy use: its impact on prices, climate, and more | Epoch AI
[p/f]
Beware the Permanent Periphery—Asterisk
[r]
Do your capabilities homework — LessWrong
[r]
AGI Ruin: A List of Lethalities - LessWrong
[p/f]
A Safe Path to Open Weights - Thinking Machines Lab
[r]
Investigating three real-world incidents in our cybersecurity evaluations \ Anthropic
[s-i]
Design tips | Malaika Aiyar
[r]
How independent researchers could investigate AI propensities after misalignment incidents - METR
[r]
Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs
[p/f]
Our position on open-weights models \ Anthropic
[r]
Introducing MAI-Cyber-1-Flash inside MDASH | Microsoft AI
[r]
What makes a good monitoring prompt? – Apollo Research
[p/f]
The OpenAI/Hugging Face Incident: Challenges in Controlling and Containing Cyber-Capable AI Systems — Institute for AI Policy and Strategy
[r]
Metrics of Agent Ability - METR
[r]
The Long (Self-)Correction — LessWrong
[s-i]
First, Make Me Care, by Gwern · Gwern.net
[r]
Séb Krier on X: "The way people discuss AI incidents is pretty important and I'm a bit concerned we're sleepwalking into a bad world littered with bad abstractions. How you label something is often a lossy compressions of a causal model, so the words you use affects which hypotheses people update https://t.co/IsMLwiMhR4" / X
[r]
Discovering Concept-Editing Algorithms With LLM Agents
[r]
Does distilling Claude carry the persona with it? — LessWrong
[r]
Rule Thinkers In, Not Out — LessWrong
[r]
Introducing Claude Opus 5 \ Anthropic
[r]
Are we existentially threatened by the type of AI misalignment seen in the OpenAI Hugging Face attack? — LessWrong
[r]
Announcing AIXI Labs — LessWrong
[r]
How our new Control Red Team is stress-testing frontier monitors | AISI Work
[r]
Your AIs don't do what you want. This is really bad
[r]
OpenAI and Hugging Face partner to address security incident during model evaluation | OpenAI
[r]
Security incident disclosure — July 2026
[p/f]
EXCLUSIVE: US, China to hold AI talks in September, sources say | Reuters
[r]
Tencent Hy
[r]
Differential acceleration of alignment-relevant capabilities is a bad bet — LessWrong
[r]
Coding vs thinking — Paradigm 3
[r]
Paradigm 3
[r]
What we’d like to fund — Paradigm 3
[r]
Safety and alignment in an era of long-horizon models | OpenAI
[p/f]
Don't default to nonprofit - by Carol and Austin Chen
[r]
which_claude_is_k3/writeups/write_up.md at main · rgreenblatt/which_claude_is_k3
[c]
China's Technological Playbook - by Sophie Kim
[p/f]
Judd Rosenblatt on X: "Xi Jinping AI speech transcript: Distinguished colleagues and guests, ladies and gentlemen, friends, 70 years ago, a group of young scholars proposed the concept of artificial intelligence for the first time at the Dartmouth workshop in New Hampshire of the United States. In" / X
[r]
[2607.14345] Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values
[r]
We need 3rd party Training-Run Assessments — LessWrong
[r]
When AI builds itself \ Anthropic
[r]
RL Post-Training on Macs | Pluralis Research
[r]
Inkling: Our Open-Weights Model - Thinking Machines Lab
[r]
Can risk aversion learned at low stakes generalize to astronomically high stakes?
[p/f]
How I think about catastrophic biological risk (part II): risk breakdown by type of prevention
[p/f]
How I think about catastrophic biological risk (part I): risk breakdown by type of response
[p/f]
The Four Pillars: A Hypothesis for Countering Catastrophic Biological Risk
[c]
The Dangers of Mirrored Life
[s-i]
‘Give Away Your Legos’ and Other Commandments for Scaling Startups | First Round Review
[r]
Toward A Public Science of Model Behavior | Transluce AI
[s-i]
Policy Memo : Broad Institute of MIT and Harvard
[r]
Notes on Inference Integrity - by James Tillman - ForeWord
[p/f]
The current bottleneck is political will, not research — LessWrong
[c]
The missing half of AI futurism debates
[p/f]
What we learned from 1,604 Chinese AI job postings
[r]
Frontier labs don’t use most AI compute (yet) - by Josh You
[r]
Selective Optimism: a critique of AI 2040 — LessWrong
[p/f]
AI-Enabled Coups: How a Small Group Could Use AI to Seize Power | Forethought
[c]
Séb Krier on X: "Concepts of a Plan" / X
[p/f]
Total research transparency would be nice - by Ajeya Cotra
[r]
To be legible, evidence of misalignment probably has to be behavioral
[s-i]
Rest in motion
[s-i]
Do Things that Don't Scale
[s-i]
Humans are not automatically strategic — LessWrong
[s-i]
A "Failure to Evaluate Return-on-Time" Fallacy — LessWrong
[r]
I think alignment work is more promising than control work — LessWrong
[s-i]
Be impatient | benkuhn.net
[s-i]
Shut up and do the impossible! — LessWrong
[s-i]
Relentlessly Resourceful
[s-i]
What We Look for in Founders
[r]
The Case for Model Forensics — LessWrong
[r]
Alignment pretraining could backfire — LessWrong
[p/f]
The Invisible Side of AI Governance — LessWrong
[c]
AI catastrophe: more like a genocide than a thought experiment — LessWrong
[r]
Why are adversaries assumed to be incapable of responding to AI risk? — LessWrong
[r]
Optimisation over non-stationary distributions creates weirder minds — LessWrong
[r]
A basic systems architecture for AI agents that do autonomous research — LessWrong
[r]
Efficient tradeoffs and the safety-usefulness tradeoff model — LessWrong
[s-i]
How to be More Agentic - by Cate Hall - Useful Fictions
[r]
Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models
[p/f]
Why I don't persuade people to do AI safety
[r]
Bad Problems Don't Stop Being Bad Because Somebody's Wrong About Fault Analysis — LessWrong
[s-i]
how to write an essay | scribbles in the margins
[r]
secret-loyalties-whitepaper.pdf
[r]
The behavioral selection model for predicting AI motivations — LessWrong
[r]
Your Left Brain Doesn't Trade With Your Right — LessWrong
[p/f]
escaping flatland: career advice for CS undergrads
[c]
It's nice of you to worry about me, but I really do have a life — LessWrong
[p/f]
How to Actually Spend Billions on AI Safety - by Sophie Kim
[p/f]
A playbook for field strategy - by Dewi Erwan
[r]
Natural Language Autoencoders \ Anthropic
[r]
Risk from fitness-seeking AIs: mechanisms and mitigations
[r]
Not a Paper: "Frontier Lab CEOs are Capable of In-Context Scheming" — LessWrong
[p/f]
Not All Compute is Created Equal
[r]
Fail safe(r) at alignment by channeling reward-hacking into a "spillway" motivation
[s-i]
Half A Month Of Consolation Writing Advice
[c]
The Orange - The Gladdest Thing
[s-i]
An extremely non-comprehensive list of how to increase your surface area for luck and magic (and instantly sprinkle fairy dust on your life)
[r]
From personas to intentions: towards a science of motivations for AI models — LessWrong
[c]
Annoyingly Principled People, and what befalls them — LessWrong
[r]
Hidden Role Games as a Trusted Model Eval - James Lucassen's Blog
[r]
Model organisms researchers should check whether high LRs defeat their model organisms — LessWrong
[r]
Steering Might Stop Working Soon — LessWrong
[s-i]
Do Thing, Do One Thing
[p/f]
What 3,654 Job Postings Tell Us About Talent Needs in AI Safety — EA Forum
[r]
If Mythos actually made Anthropic employees 4x more productive, I would radically shorten my timelines
[c]
Why You Can’t Just Do Things - Octopusyarn
[r]
Anthropic repeatedly accidentally trained against the CoT, demonstrating inadequate processes
[p/f]
AI Populism's Warning Shots - by Jasmine Sun
[s-i]
How to walk through walls - by Henrik Karlsson
[r]
[2603.20639] Agentic AI and the next intelligence explosion
[s-i]
A woefully incomplete guide to technical upskilling
[r]
An Apple-Picking Model of AI R&D | Tom Cunningham – Tom Cunningham
[c]
11 pieces of advice for children — LessWrong
[p/f]
The state of AI safety in four fake graphs — LessWrong
[r]
Product Alignment is not Superintelligence Alignment (and we need the latter to survive) — LessWrong
[c]
Academic Proof-of-Work in the Age of LLMs — LessWrong
[r]
Fitness-Seekers: Generalizing the Reward-Seeking Threat Model
[p/f]
AI Safety Talent Needs in 2026: Insights for Field-Building Organizations
[c]
Facing the Precipice of History - Chanden Climaco
[p/f]
AI Safety Needs Startups - by Joshua Landes and LTM
[s-i]
Your Work Will Change You Whether You Like It Or Not
[s-i]
Strategic Taste
[r]
Are AIs more likely to pursue on-episode or beyond-episode reward?
[p/f]
There should be ‘general managers’ for more of the world’s important problems
[p/f]
We don't need more founders in AI safety - by Gauraventh
[p/f]
Lessons from a year of university AI safety field building — LessWrong
[r]
Separating Prediction from Goal-Seeking — LessWrong
[p/f]
Two Skillsets You Need to Launch an Impactful AI Safety Project — LessWrong
[r]
How to Design Environments for Understanding Model Motives — LessWrong
[r]
Martian Interpretability Challenge: The Core Problems In Interpretability — LessWrong
[r]
Prefill awareness: can LLMs tell when “their” message history has been tampered with? — LessWrong
[c]
Don't Let LLMs Write For You — LessWrong
[c]
Tell Culture — LessWrong
[r]
How to win a best paper award (or, an opinionated take on how to do important research)
[r]
Current activation oracles are hard to use — LessWrong
[p/f]
The current SOTA model was released without safety evals — LessWrong
[r]
Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers — LessWrong
[p/f]
Good Ideas Aren't Enough in AI Policy - Andrew Wei
[r]
Instant LLM Updates with Doc-to-LoRA and Text-to-LoRA
[p/f]
The Stakes of Our Work - by Celeste Li - heart of hearts
[p/f]
Why AI won’t go well unless sensible people like you speak up and act.
[r]
Persona Parasitology — LessWrong
[r]
Value systematization: how values become coherent (and misaligned) — AI Alignment Forum
[r]
The Persona Selection Model: Why AI Assistants might Behave like Humans
[c]
The 2026 Global Intelligence Crisis - Citadel Securities
[r]
Questionable practices in machine learning
[r]
Managed vs Unmanaged Agency — LessWrong
[r]
Mapping LLM attractor states — LessWrong
[r]
METR's 14h 50% Horizon Impacts The Economy More Than ASI Timelines — LessWrong
[c]
Changing the world for the worse — LessWrong
[c]
You don't create a culture – Signal v. Noise
[c]
Minimal-trust investigations
[r]
Alignment to Evil — LessWrong
[p/f]
If you don't feel deeply confused about AGI risk, something's wrong — LessWrong
[r]
Aligning to Virtues — LessWrong
[r]
Did Claude 3 Opus align itself via gradient hacking? — LessWrong
[s-i]
My six stages of learning to be a socially normal person
[c]
Good conversations have lots of doorknobs
[c]
21 Facts About Throwing Good Parties
[s-i]
How to Manage Relationships Like a Psychopath
[s-i]
Most of Your Efforts are Wasted. Here’s the Framework to Fix It.
[r]
Will reward-seekers respond to distant incentives? — LessWrong
[p/f]
Two Buckets - by Gauraventh - Dhaniya
[s-i]
How I've run major projects | benkuhn.net
[r]
Frontier Safety Framework Report - Gemini 3 Pro (November, 2025) v2
[c]
Status Is The Game Of The Losers' Bracket — LessWrong
[c]
Where is the Capital? An Overview — LessWrong
[r]
Omniscaling to MNIST — LessWrong
[r]
Legible vs. Illegible AI Safety Problems — LessWrong