NeuroQwen gives every Qwen layer simulated neurotransmitters with PID homeostasis
Crazy-Bat8405 · reddit · 2026-11-12
A developer built NeuroQwen, a Colab notebook + GitHub repo that runs 4-bit Qwen2.5 with a simulated neurochemical system on top — no weight edits.
- Mechanism: each transformer layer keeps a live vector of 6 "neurotransmitters" (dopamine, serotonin, norepinephrine, acetylcholine, glutamate, GABA); on every token, random noise perturbs levels, a PID controller pulls them back to balance (homeostasis), opposing chemicals inhibit each other, and chemistry diffuses between neighboring layers.
- Push/pull prompts: random "environment" prompts shove a chemical (e.g. "everything feels urgent"), and the controller writes counter-prompts against the biggest deviation.
- Effects: glutamate/GABA scales MLP output, acetylcholine scales attention gain, dopamine boosts attention output and logit confidence, serotonin damps attention and narrows top-p, norepinephrine adds activation noise and raises temperature — with depth-dependent sensitivities.
The author is seeking feedback on better chemical-to-computation mappings, training it as a chemistry-gated LoRA instead of inference-time hooks, and cleaner metrics for whether homeostasis affects output quality.
More from Research
- ICLR abstract word-frequency shift: 'novel' and 'comprehensive' plummet as LLM fingerprints fade — y_m_asano · 2026-11-12
- Evolving Infra: agent-adaptive LLM serving cuts latency up to 4.1x vs NVIDIA Dynamo — mit-han-lab · 2026-11-12
- Isomorphic Labs President on Using AI to Turn Drug Discovery into Precision Engineering — No Priors · 2026-11-12
- CreativityPrism benchmark scores 17 frontier and open LLMs across 8 creative tasks — DanielKhashabi · 2026-11-12
- KQuant: open-source MLX quants cut Qwen KL divergence up to 3.7x vs public 4-bit — m4rkGS · 2026-11-12
- Possible eval leak: all frontier models fail a GPQA Diamond question whose official answer seems wrong — qberthet · 2026-11-12