LLM-assisted neuron simulations reportedly revisit earlier path-integration hypotheses
The person behind the two-day effort says prior research makes the results “less discovery” and more simulations of ideas already proposed by other scientists.
TLDR
The author of a two-day effort to push automated scientific discovery with LLMs says a neuroscientist pointed them to earlier literature identifying hΔ neurons as likely candidates for path integration, plus dissertations reporting experiments on those neurons. The author now describes the work as simulations of earlier hypotheses, while believing the ideas were simulated independently through unbiased screens of neural wiring. They say embargoed abstracts still make it difficult to assess whether path integration occurs through synaptic updates, persistent activity or both.
Combined views
2
1 Source, first seen 15d ago