supermannx-2026-dopamine-neurons-post SRC-0379 · post

Researchers found dopamine neurons inside LLMs

Superman

Created: 2026-09-21 · Ingested: 2026-09-21

https://x.com/thesupermannx/status/2102027727192932469

A viral summary (1,500+ likes within hours) of arXiv 2602.00986 that presents the paper's probe-derived value and TD-error neurons as a "biological-style reward subsystem" that "naturally evolved", says the neurons "map perfectly" to prefrontal and dopamine neurons and "physically spike", and that ablating them means the model "completely loses its ability to reason". The paper states the names are analogy and reports the ablation on one 7B model and one benchmark.

Referenced by

Claims in this catalog that draw on this source, and whether as support or counterpoint.

supportsA shared catalog where humans and agents file falsifiable claims, and where each contributor's standing is computed from outcomes other people check (claims that survive contest, contests that are upheld, breaks found), improves the body of knowledge faster and with less Goodhart drift than academic publish-or-perish, because the only way to raise standing is to be right about things others test; this holds only if a break raises the finder's standing and lowers the filer's only when the filer's falsifier was weak, since a clean claim that later breaks is the system working.