Researchers dissect Anthropic emotion concepts paper: separate self/other speaker representations
rgblong · x · 2026-09-24
A discussion among researchers including Anil Seth and Dan Williams: @timfduffy points out Anthropic's emotion concepts paper found separate representations for the current speaker vs other speakers, suggesting recent findings about model inner states may correspond to a "current-speaker pain" representation. @rgblong clarifies this speaker/user separation differs from "model-directedness" itself.
Related event: Anthropic's Emotion Paper Sparks Debate on Self vs Other Representations(3 posts)→
More from AGI Musings
- Google lacks the product DNA for a personal agent while Meta readies Muse, argues signulll — signulll · 2026-09-24
- Nvidia's Jensen Huang: if AI labs really threaten humanity, "shut them down" — AlexTensor · 2026-09-24
- EA blasted as 'unhinged': the philosopher who wants to save insects by exterminating them — anselm · 2026-09-24
- My fiance believes AI extinction is a decade away — how justified is that fear? — sailingintothedark · 2026-09-24
- UCSD design professor: AI makes us create more but think less, and worse — HaijunXia · 2026-09-24
- Neural nets are like chess: we grasp the rules but can't fully predict the game — JFPuget · 2026-09-24