Episode
Ep 255: Does this research explain how LLMs work?
- Podcast
- ToKCast
- Published
- Jan 14, 2026
- Duration seconds
- 4965
- Processing state
not_requested
Actions
POST https://stenobird.com/v1/public/podcasts/tokcast-76992/episodes/ep-255-does-this-research-explain-how-llms-work/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/tokcast-76992/ep-255-does-this-research-explain-how-llms-work.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
I take a look at these three papers: 1. https://www.arxiv.org/abs/2512.22471 2. https://arxiv.org/abs/2512.23752 3. https://arxiv.org/abs/2512.22473 Collectively titled "The Bayesian Attention Trilogy" along with some other material - in particular an interview with one of the authors "Vishal Misra" - https://www.engineering.columbia.edu/faculty-staff/directory/vishal-misra For those familiar with my output on this you can probably skip to about halfway through at 42:40. Prior to this is a lot of background on Induction, Bayesianism, Critical Rationalism and so on that people may have heard from me before in different contexts - although for what it's worth these are new ways of expressing those ideas. At the end I am reacting to a video found here: https://www.youtube.com/watch?v=uRuY0ozEm3Q