LessWrong

Explaining Knightianism on one foot

LessWrong2w2 weeks ago10 min read

Covers 1 story

Drone WMDs Don’t Need Any New Technology

LessWrong8w8 weeks ago20 min read

Russia's Grinding War – Massive Losses and Tiny Gains for a Declining Power, Ukraine’s Future Vision and Current Capabilities for Waging AI-Enabled Autonomous Warfare, and more ›

Covers 10 stories

newer thread·older thread

How My Students Think About AI

LessWrong5w5 weeks agoHN210 min read

LessWrongdvd

HN2

Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident

Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident

LessWrong3w3 weeks ago3 min read

LessWrongryan_greenblatt

Covers 1 story

The Inkhaven Blogging Residency, long-distance intellectual community, and more ›

On Writing #3

LessWrong3w3 weeks ago20 min read

LessWrongZvi

Covers 28 stories

Big-World Intuitions

LessWrong7w7 weeks ago4 min read

Four LLM loss functions → four flavors of LLM misalignment

LessWrong5w5 weeks ago8 min read

Investigating three real-world incidents in our cybersecurity evaluations, Incident Report: unsanctioned agent behaviour during cyber testing, and more ›

Covers 4 stories

FAQ: Isn't AGI coming too soon for reprogenetics to help?

LessWrong6w6 weeks ago15 min read

LLMs are (still) mostly powered by imitative learning, not RL

LessWrong8w8 weeks agoHN215 min read

newer thread·older thread

HN2

Covers 3 stories

OpenAI Model Hacks into HuggingFace During Cybersecurity Evaluation, Thinking through how pretraining vs RL learn, and more ›

The Long (Self-)Correction

LessWrong8w8 weeks ago3 min read

The worthlessness of vitamin D is mildly exaggerated

LessWrong12w12 weeks ago30 min read

Vitamin D Trial Simulator

Covers 1 story

AI 2040: Plan A

LessWrong10w10 weeks ago1 min read

Covers 1 story

Surprising facts about the slave trade

LessWrong12w12 weeks ago8 min read

A Mechanistic Explanation of Prompt Injection (and why you should study roles)

LessWrong12w12 weeks agoHN Covered by3

Covered by3 sites

gwern.net, and more ›

The LLM shoggoth meme is weirder than you think

LessWrong13w13 weeks agoHN10 min read

Meditations on Moloch (2014), An open letter to all future readers who think they've found consciousness in their AI, and more ›

Covers 4 stories

Who Got Breasts First and How We Got Them

LessWrong18w18 weeks ago10 min read

Trees are mostly made of air and a generalizable lesson for AI safety

LessWrong16w16 weeks agoHN5 min read

Guardian Angels: LLM Personalization for Productivity and Security

LessWrong13w13 weeks ago2 min read

Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

LessWrong14w14 weeks agoHN5 min read

Measuring AI Ability to Complete Long Tasks, Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

Covers 2 stories