Daniel Filan
@dfrsrchtwts
Followers
2K
Following
842
Media
101
Statuses
1K
Research manager at MATS. Want to usher in an era of human-friendly superintelligence, don't know how. Podcast: https://t.co/gM752yhlTd
Joined June 2020
active inference baby.
ChatGPT kept telling people you could could import tabs from ascii at their website so they just built a feature to do that
0
0
6
RT @jessi_cata: There has been much criticism of the AI 2027 model. As a check, I ran a Monte Carlo model based on METR data (2032 median).….
0
5
0
RT @AXRPodcast: My apologies: if you downloaded my most recent episode, my audio cut out around 0:57:40. The issue should be fixed if you r….
0
1
0
New episode with @SamuelAlbanie, where we discuss the recent Google DeepMind paper "An Approach to Technical AGI Safety and Security"! Link to watch below.
Episode 45 - Samuel Albanie on DeepMind's AGI Safety Approach.
1
4
26
Ways you could do this:.- do a bunch of safety testing on un-RL'd model, if it's basically OK, deploy RL'd model.- mostly deploy RL'd model, but 0.1% of the time use the un-RL'd model instead, halt and catch fire if you see anything too bad in that 0.1%.
1
0
4
This is probably a dumb question, but what's the problem with doing RL on CoT for safety? People are like "oh it makes the CoT less interpretable so you can't notice problems" but can't you keep the un-RL'd model around and see if it has issues also?.
5
1
12
New episode with @petersalib! We chat about how giving AIs rights might make them less inclined to take over, and also why legal academia is so weird. Link to watch in reply.
Episode 44 - Peter Salib on AI Rights for Human Safety.
1
2
18
RT @austinc3301: 🚀 We're launching mentor applications for SPAR's Fall 2025 round!. @SPARexec is a part-time, remote research program where….
0
2
0
RT @Jsevillamol: A couple of weeks ago I posted a summary of Epoch's mission, clearing up some common misunderstanding of what we are tryin….
0
5
0
RT @OwainEvans_UK: Some recent talks/interviews:.Podcast on introspection, self-awareness and emergent misalignment .
0
10
0
RT @binarybits: Something that comes through clearly in the DeepSeek R1 research paper, and I wish was more broadly understood, is that the….
0
18
0
New episode with @davlindner, covering his work on MONA! Check it out - video link in reply.
Episode 43 - David Lindner on Myopic Optimization with Non-myopic Approval.
1
4
30
RT @MariusHobbhahn: LLMs Often Know When They Are Being Evaluated!. We investigate frontier LLMs across 1000 datapoints from 61 distinct da….
0
81
0
New episode with @OwainEvans_UK! Covers work on emergent misalignment and more!
Episode 42 - Owain Evans on LLM Psychology.
2
9
73
RT @safe_paper: Large Language Models Often Know When They Are Being Evaluated.Joe Needham, Giles Edkins (@gdedkins), Govind Pimpale (@Govi….
0
23
0