Site replays Qwen3-4B neurons in 3D with sound, showing how it confidently misspells then fails
silvia_stocchi · reddit · 2026-11-18
A free, no-signup site (watchaithink.caby.dev) replays a real small model (Qwen3-4B, 36 layers) answering real questions, rendering neuron activity in 3D with sound — steady tone for confident answers, wavering for unsure ones. In one replay the model, asked to reverse "brainstorm", misspells it in its private thinking chain as "mortsniarb" and confidently reverses its own typo. The author notes these are six pre-recorded replays, not a live model, showing what happened rather than why, and say nothing about ChatGPT or Claude internals; inspired by lak7's thinkspace (live GPT-2).
More from Fun
- Apple Watch bricks after one shower while a $15 Keep band survives many — oran_ge · 2026-11-22
- Surreal "Overthinking" illustrations show a writer swarmed by snakes and eyes — gizakdag · 2026-11-22
- Cinematic UFO folklore music video built with Suno and SLOPCANNON — DarkroomSignal · 2026-11-22
- How to spot a model distilled from Astra: it can't stop drawing armillary spheres — teortaxesTex · 2026-11-22
- Yudkowsky's 'rolling cylinder' defense draws fire: 'too many sycophants around you' — teortaxesTex · 2026-11-22
- Big Tech PM directors preach 'AI-era PM' on LinkedIn; ex-PM calls it their swan song — nikitabier · 2026-11-22