Interpretability researchers clash: is it a V neuron, and what does SAE failure to disentangle mean?
livgorton · x · 2026-10-22
A technical interpretability debate between livgorton and another researcher over whether a neuron is a "V neuron" and the role of superposition:
- livgorton notes the essay strongly claims it's a V neuron — the very claim people are responding to — and asks the author to formalize the actual claim to avoid talking past each other
- The counterpoint: if features were grouped due to a relevant overlapping property, a well-trained SAE would pull out that property; SAEs failing to disentangle is evidence against superposition
- livgorton replies that recursive generalisation is fully compatible with superposition, and what gets superposed together over training is itself studyable
A representative clash over what SAE evidence does and doesn't prove in mechanistic interpretability.
More from Research
- 130K-sample RAG claim verification dataset lands on Hugging Face — MaziyarPanahi · 2026-10-22
- Culture is the moat: Scobleizer on cultural data for humanoid robot learning — Scobleizer · 2026-10-22
- Tencent Hunyuan's WorkForge builds 16.7K verifiable real-world environments for work agents — TencentHunyuan · 2026-10-22
- Gauss vs the machine: what the 1-to-100 sum trick says about ML and computation — tobias_rees · 2026-10-22
- Radiolab's 'Math Vs Machine': OpenAI claims a Millennium Prize Problem solve — stevenstrogatz · 2026-10-22
- UK/Nigeria driver-coaching NLG study wins Best Evaluation award at INLG 2026 — EhudReiter · 2026-10-22