Out of everything we covered on
@thursdai_pod today (Claude Code leak, SessionGate,
@googlegemma 4, 1bit quantization) this was the most insane.
Please read this research from Anthropic, they are leading the world in Mech Interp (aka, LLM brain surgery) and this is just fascinating!
They were able to "prove" that the LLMs have emotions, or at least, the characters that the LLMs play, have emotions. They were able to distill them, and clamp or hyper those emotions up, to show difference in outcomes.
For example the way the model feels frustration while doing an impossible coding task, and then feels "relief" when it's finally done (though cheated through it) is just fascinating.
New Anthropic research: Emotion concepts and their function in a large language model.
All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.