AI Research Circle [members and +1s]
Monday · 2025-12-08 · 7:30 PM - 8:45 PM
We examined how linear directions in a model's activation space correspond to specific character traits like sycophancy, hallucination propensity, and moral reasoning. The group discussed what personality means for language models and the implications for alignment, safety, and developer tools. Paper: "Persona Vectors" (Chen et al., 2025).
More like this
Personal agents handling your credentials and calendar need serious security architecture. This week's discussion covers Meta's Muse approach, sandboxing, permission isolation, and approval flows, plus the harder question of what actually makes an agent trustworthy enough to run unsupervised.
A casual roundtable on cutting-edge LLMs, robotics, simulation, and BCI research with folks actively pushing the field forward. Weekly robotics discussions continue here, plus free pizza and a low-fi DJ setting.
OtherMinds Lab explores whether life could evolve through disordered proteins, extremophile survival, or silicon-based chemistry by examining three research papers on alternative biochemistry. A moderated discussion diving into what actually defines life and how we'd recognize it beyond Earth, no specialist background required.
Want more like this?
Get weekly picks matched to what you're into.