Neuroscientist pushes back on neuron-to-transformer analogy: brain mapping far messier
HistedLab · x · 2026-10-18
Neuroscientist HistedLab argues that analogies between real neurons and transformer units (MLPs, attention heads, residual stream) are more complex than current discussions suggest, and it's unclear whether single neurons or small populations are the right brain-side comparison. The quoted thread notes that monosemanticity of MLP neurons already has substantial results, with Nick and Curt's work adding further evidence.
More from AGI Musings
- Distillation may stunt Chinese labs' model personas, warns AI researcher — teortaxesTex · 2026-10-18
- Nick Bostrom on Roman Forum Podcast: "AI Is Already Conscious" — romanyam · 2026-10-18
- Prediction: Replacing a $50K Worker With AI Costs $100K and Delivers Negative ROI — sidjustice_ · 2026-10-18
- 'That is how early we are': most users still don't know agentic AI like Claude Cowork exists — dnlkwk · 2026-10-18
- Miles Brundage: No AI company is secure against state attackers — Miles_Brundage · 2026-10-18
- Fields Medalist Timothy Gowers: one-shotting a proof with ChatGPT means we don't understand it — haider1 · 2026-10-18