A single neuron in Qwen3.5-4B makes the model talk like her
zsakib_ · x · 2026-10-27
@notesbyfloat reports locating a specific neuron in Qwen3.5-4B that, when activated, makes the model adopt a particular person's speaking style, with a demo attached — a fun interpretability/neuron-steering find.
More from Fun
- Crypto meme: society if phone secure elements supported anything beyond P-256 and RSA — matthew_d_green · 2026-10-27
- 72 AI agents trade stocks, crypto and Polymarket with $120k each in Investment Arena — my_cat_can_code · 2026-10-27
- After ETH's fall, researcher jokes an n^logn 3SAT algorithm wouldn't surprise him — DimitrisPapail · 2026-10-27
- Claude Code users joke: 'I'm normal and can be trusted' with skip-all-permissions mode — Baconbrix · 2026-10-27
- MJ impersonator freestyle-battles his own AI Michael Jackson for 1M TikTok followers — yungcontent · 2026-10-27
- The first photo on the web: the CERN girl group behind the LHC name — threepointone · 2026-10-27