Show Notes
Earlier this year, the paper "Emergent Misalignment" made the rounds on AI x-risk social media for seemingly showing LLMs generalizing from 'misaligned' training data of insecure code to acting comically evil in response to innocuous questions. In this episode, I chat with one of the authors of that paper, Owain Evans, about that research as well as other work he's done to understand the psychology of large language models. Patreon: https://www.patreon.com/axrpodcast Ko-fi: https://ko-fi.com/axrpodcast Transcript: https://axrp.net/episode/2025/06/06/episode-42-owain-evans-llm-psychology.html
Topics we discuss, and timestamps: 0:00:37 Why introspection? 0:06:24 Experiments in "Looking Inward" 0:15:11 Why fine-tune for introspection? 0:22:32 Does "Looking Inward" test introspection, or something else? 0:34:14 Interpreting the results of "Looking Inward" 0:44:56 Limitations to introspection? 0:49:54 "Tell me about yourself", and its relation to other papers 1:05:45 Backdoor results 1:12:01 Emergent Misalignment 1:22:13 Why so hammy, and so infrequently evil? 1:36:31 Why emergent misalignment? 1:46:45 Emergent misalignment and other types of misalignment 1:53:57 Is emergent misalignment good news? 2:00:01 Follow-up work to "Emergent Misalignment" 2:03:10 Reception of "Emergent Misalignment" vs other papers 2:07:43 Evil numbers 2:12:20 Following Owain's research
Links for Owain: Truthful AI: https://www.truthfulai.org Owain's website: https://owainevans.github.io/ Owain's twitter/X account: https://twitter.com/OwainEvans_UK
Research we discuss: Looking Inward: Language Models Can Learn About Themselves by Introspection: https://arxiv.org/abs/2410.13787 Tell me about yourself: LLMs are aware of their learned behaviors: https://arxiv.org/abs/2501.11120 Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data: https://arxiv.org/abs/2406.14546 Emergent Misalignment: Narrow fine-tuning can produce broadly misaligned LLMs: https://arxiv.org/abs/2502.17424 X/Twitter thread of GPT-4.1 emergent misalignment results: https://x.com/OwainEvans_UK/status/1912701650051190852 Taken out of context: On measuring situational awareness in LLMs: https://arxiv.org/abs/2309.00667
Episode art by Hamish Doodles: hamishdoodles.com
More from AXRP - the AI X-risk Research Podcast
Remember AI 2027? Not AI 2040, the newest coolest thing AI Futures Project has done, but AI 2027, their OG product? At long last, we have an AXRP episode about it. Enjoy!
How does game theory work when everyone is a computer program who can read everyone else's source code? This is the problem of 'program equilibria'. In this episode, I talk with Caspar Oesterheld on work he's done on equilibria of programs that simulate each other, and how robust these equilibria are. Patreon: https://www.patreon.com/axrpodcast Ko-fi: https://ko-fi.com/axrpodcast Transcript: https://axrp.net/episode/2026/02/18/episode-49-caspar-oesterheld-program-equilibrium.html Note from Caspar on 2:00:06: At least given my current interpretation of what you say here, my answer is wrong. What actually happens is that we're just back in the uncorrelated case. Basically my simulations will be a simulated repeated game in which everything is correlated _because I feed you my random sequence_ and your simulations will be a repeated game where everything is correlated. Halting works the same as usual. But of course what we end up actually playing will be uncorrelated. We discuss something like this later in the episode.
In this episode, Guive Assadi argues that we should give AIs property rights, so that they are integrated in our system of property and come to rely on it. The claim is that this means that AIs would not kill or steal from humans, because that would undermine the whole property system, which would be extremely valuable to them. Patreon: https://www.patreon.com/axrpodcast Ko-fi: https://ko-fi.com/axrpodcast Transcript: https://axrp.net/episode/2026/02/15/episode-48-guive-assadi-ai-property-rights.html
When METR says something like "Claude Opus 4.5 has a 50% time horizon of 4 hours and 50 minutes", what does that mean? In this episode David Rein, METR researcher and co-author of the paper "Measuring AI ability to complete long tasks", talks about METR's work on measuring time horizons, the methodology behind those numbers, and what work remains to be done in this domain. Patreon: https://www.patreon.com/axrpodcast Ko-fi: https://ko-fi.com/axrpodcast Transcript: https://axrp.net/episode/2026/01/03/episode-47-david-rein-metr-time-horizons.html
Could AI enable a small group to gain power over a large country, and lock in their power permanently? Often, people worried about catastrophic risks from AI have been concerned with misalignment risks. In this episode, Tom Davidson talks about a risk that could be comparably important: that of AI-enabled coups. Patreon: https://www.patreon.com/axrpodcast Ko-fi: https://ko-fi.com/axrpodcast Transcript: https://axrp.net/episode/2025/08/07/episode-46-tom-davidson-ai-enabled-coups.html