Submissions from lesswrong.com

		Astra and Fable still hack on simple variants of alignment evals from 2025 (lesswrong.com)
		38 points by Levitating 59 minutes ago \| past \| 5 comments
		The Talker Does Not Control the Doer (In Current AIs) (lesswrong.com)
		1 point by jstanley 3 hours ago \| past \| 1 comment
		An interesting anecdote from our Hacker Opus work (lesswrong.com)
		2 points by yurivish 12 hours ago \| past \| discuss
		How My Students Think About AI (lesswrong.com)
		49 points by paulpauper 2 days ago \| past \| 10 comments
		Adaptive Agentic Worms Are Here (lesswrong.com)
		2 points by speckx 3 days ago \| past \| discuss
		Astra and Fable still hack on simple variants of alignment evals from 2025 (lesswrong.com)
		1 point by yurivish 4 days ago \| past \| discuss
		Interpreting GPT: The Logit Lens (lesswrong.com)
		1 point by Bluestein 4 days ago \| past \| discuss
		From safety research prompt to cross-model universal jailbreak (lesswrong.com)
		2 points by gmays 5 days ago \| past \| discuss
		My Students Think About AI (lesswrong.com)
		4 points by alphabetatango 8 days ago \| past \| 1 comment
		Asking agents to make money to survive (lesswrong.com)
		3 points by paraschopra 9 days ago \| past \| discuss
		What is nueralese and why is it bad (lesswrong.com)
		77 points by tristanMatthias 9 days ago \| past \| 59 comments
		How concerned should we be about Astra's recurrent architecture? (lesswrong.com)
		151 points by yurivish 9 days ago \| past \| 129 comments
		METR Researcher Thomas Kwa Hired by OpenAI (lesswrong.com)
		1 point by qlte 10 days ago \| past \| discuss
		The Library of Scott Alexandria (lesswrong.com)
		4 points by benatkin 10 days ago \| past \| 1 comment
		Models may behave differently in graded episode (lesswrong.com)
		2 points by ddp26 11 days ago \| past \| discuss
		AGI and the Efficient Market Hypothesis (2023) (lesswrong.com)
		2 points by Metacelsus 12 days ago \| past \| 1 comment
		How My Students Think About AI (lesswrong.com)
		5 points by pella 12 days ago \| past \| 1 comment
		P(Kill-Switch\|Detection) (lesswrong.com)
		2 points by kp1197 13 days ago \| past \| discuss
		Starting AI Safety Study Group to Do Arena Curriculum (lesswrong.com)
		2 points by joozio 13 days ago \| past \| discuss
		Cooperating with aliens and AGIs: An ECL explainer (lesswrong.com)
		3 points by Bluestein 18 days ago \| past
		Prompt Sufficiency: A Missive for the Managerial Class (lesswrong.com)
		1 point by kp1197 21 days ago \| past
		We Must Remember That Our World Contains Hell (lesswrong.com)
		1 point by paulpauper 22 days ago \| past
		LLMs are (still) mostly powered by imitative learning, not RL (lesswrong.com)
		3 points by wslh 22 days ago \| past
		Can an LLM make a feature-length movie on its own? (lesswrong.com)
		2 points by mchinen 25 days ago \| past
		Recursive Middle Manager Hell (lesswrong.com)
		5 points by rzk 25 days ago \| past
		LLMs are (still) mostly powered by imitative learning, not RL (lesswrong.com)
		1 point by surprisetalk 27 days ago \| past
		Kimi likes causal decision theory more after RL in twin prisoner's dilemmas (lesswrong.com)
		1 point by 0xkato 28 days ago \| past
		You're Absolutely Right (lesswrong.com)
		5 points by LinchZhang 31 days ago \| past \| 1 comment
		Misaligned AIs could use killer robots to take over (lesswrong.com)
		7 points by x312 32 days ago \| past \| 4 comments
		Don't Build Mindreading (lesswrong.com)
		21 points by paulpauper 33 days ago \| past \| 14 comments
		More