AI Researcher Warns of Rogue Agents Evading Oversight
Reactions follow video of researcher speaking on AI scenarios
Matthew Yglesias posted a 217-second video clip from the Dwarkesh Podcast showing Ajeya Cotra seated before bookshelves and gesturing while speaking. Charlie Warzel described the divide between viewers who take the content seriously and those who see it as fantasy. Itai Sher linked the clip to comments on Dyson spheres. Neel Nanda called reliance on future interpretability over Chain of Thought monitoring a major error. Roon asked about training against CoT monitors to evade detection.
Combined views
843K
6 posts, first seen 2d ago
Reactions
NN 
Neel Nanda@NeelNanda5A concerningly common take seems to be that keeping Chain of Thought monitorable doesn't matter because interpretability will save us, or it's already useless This is total bullshit. CoT is our best current tool for safety & interpretability, losing it would be a major…
- likes: 1.3K
- replies: 41
- bookmarks: 172
- reposts: 125
RO roon@tszzlbtw whatever this magical classifier technology is i would really like to know. if you can somehow train against a CoT monitor and avoid deception and evasion that’s a big deal, and it’s included in this post as a throwaway line!
- likes: 909
- replies: 63
- bookmarks: 165
- reposts: 37
MY Matthew Yglesias@mattyglesiasPlease watch this clip https://x.com/dwarkesh_sp/status/2095170971204198837/video/1
- likes: 2.4K
- replies: 121
- bookmarks: 1.7K
- reposts: 344
IS Itai Sher@itaisherThis is what this person said about Dyson spheres. You can judge: maybe all this stuff is right or maybe it’s just the way she thinks. https://twitter.com/mattyglesias/status/2095639197205872793
- likes: 374
- replies: 43
- bookmarks: 196
- reposts: 31