Все публикации

0:38:21

309. No, LLMs are not Scheming

0:23:10

308. Frontier Models Are Capable of InContext Scheming

0:28:35

307. OpenAI Email Archives

0:29:49

306. Superintelligences Will Not Spare Earth

0:20:09

305. A Funny Feature of the AI Doomster Argument

0:36:29

303. How to prevent collusion when using untrusted models to monitor each other

0:32:39

302. AGI Safety and Alignment At Google DeepMind

1:27:19

301. CAST Discussion with Max Harms

1:30:28

300. Corrigibility As Singular Target 0 and 1

0:36:31

299. Assessing the Risk of Takeover Catastrophe from LLMs

1:51:37

297. Against the Singularity Hypothesis AND discussion with David Thorstad

0:25:46

296. LLMs for Alignment Research A Safety Priority

0:22:11

295. California Senate Bill 1047

0:20:05

294. Practices for governing agentic AI systems

0:52:35

293. pDoom is 0.95

0:39:58

290. Counting Arguments Provide No Evidence For AI Doom

2:17:33

289. Gaining Capabilities by learning the plan effect mapping

1:47:44

288. Credibly Safe AI

0:39:51

287. Existential Risk Persuasion Tournament

0:39:25

285. Imitation Learning is Probably Existentially Safe 2 (Sound improved, but not perfect)

0:39:25

285. Imitation Learning is Probably Existentially Safe 2

0:50:12

284. Imitation Learning is Probably Existentially Safe 1

0:36:27

283. AI Pause Will Likely Backfire 2

0:32:28

282. AI Pause Will Likely Backfire 1