Submissions from lesswrong.com

		Opus 4.7 Part 3: Model Welfare (lesswrong.com)
		11 points by omer_k 2 days ago \| past \| 8 comments
		Claude Code sometimes hallucinates user messages (lesswrong.com)
		2 points by cubefox 5 days ago \| past \| 2 comments
		There are only four skills: design, technical, management and physical (lesswrong.com)
		3 points by samuel246 5 days ago \| past \| discuss
		Summarizing and Reviewing my earliest ML research paper, 7 years later (lesswrong.com)
		2 points by joozio 6 days ago \| past \| discuss
		Resources for starting and growing an AI safety org (lesswrong.com)
		1 point by omer_k 6 days ago \| past \| discuss
		Only Law Can Prevent Extinction (lesswrong.com)
		3 points by namanyayg 6 days ago \| past \| 1 comment
		LLMs will soon disrupt algorithmic media feeds (lesswrong.com)
		3 points by linhns 6 days ago \| past \| discuss
		Working hurts less than procrastinating, we fear the twinge of starting (2011) (lesswrong.com)
		14 points by davikr 7 days ago \| past \| 2 comments
		The AlphaFold moment for materials is not any time soon (lesswrong.com)
		8 points by gmays 10 days ago \| past \| discuss
		Morale (lesswrong.com)
		2 points by jger15 10 days ago \| past \| discuss
		You're gonna need a bigger benchmark, METR (lesswrong.com)
		3 points by frmsaul 11 days ago \| past \| discuss
		Hypotheses for Why Models Fail on Long Tasks (lesswrong.com)
		1 point by joozio 12 days ago \| past \| discuss
		Splitting Mounjaro pens for fun and profit (lesswrong.com)
		2 points by henryaj 12 days ago \| past \| discuss
		We're running out of benchmarks to upper bound AI capabilities (lesswrong.com)
		15 points by gmays 14 days ago \| past \| 10 comments
		AIs can now do easy-to-verify SWE tasks, I've shortened timelines (lesswrong.com)
		3 points by gmays 14 days ago \| past
		The effects of caffeine consumption do not decay with a ~5 hour half-life (lesswrong.com)
		101 points by swah 15 days ago \| past \| 105 comments
		My Picture of the Present in AI (lesswrong.com)
		1 point by speckx 15 days ago \| past
		Most people can't juggle one ball (lesswrong.com)
		507 points by surprisetalk 16 days ago \| past \| 174 comments
		"Alignment" and "Safety", Part One: What Is "AI Safety"? (lesswrong.com)
		1 point by joozio 18 days ago \| past
		Paper Close Reading: "Why Language Models Hallucinate" (lesswrong.com)
		2 points by joozio 19 days ago \| past
		Estimates of the expected utility gain of AI Safety Research (lesswrong.com)
		1 point by joozio 19 days ago \| past
		What I like about MATS and Research Management (lesswrong.com)
		2 points by joozio 19 days ago \| past
		Predicting When RL Training Breaks Chain-of-Thought Monitorability (lesswrong.com)
		2 points by gmays 19 days ago \| past
		AI Safety at the Frontier: Paper Highlights of February and March 2026 (lesswrong.com)
		2 points by joozio 20 days ago \| past
		How to emotionally grasp the risks of AI Safety (lesswrong.com)
		3 points by joozio 21 days ago \| past
		You can't imitation-learn how to continual-learn (lesswrong.com)
		2 points by paulpauper 21 days ago \| past
		A Mirror Test for LLMs (lesswrong.com)
		2 points by gmays 22 days ago \| past
		I'm Suing Anthropic for Unauthorized Use of My Personality (lesswrong.com)
		5 points by usrme 23 days ago \| past \| 2 comments
		Why did everything take so long? (lesswrong.com)
		2 points by jstanley 24 days ago \| past
		The state of AI safety in four fake graphs (lesswrong.com)
		3 points by allenleee 24 days ago \| past
		More