Residual-stream ontologizer clusters 1,711 EA Global attendees into 48 types
amplifiedamp · x · 2026-10-25
To decide whom to prioritize at EA Global, amplifiedamp of WakaruLabs built a color-coded "ontologizer" using residual stream activations from Trinity Large TrueBase 400B (layer 59) plus their own library, clustering all 1,711 NYC attendee profiles.
The resulting tree maps 48 attendee archetypes (8 heads × 6 entries), each auto-described by Claude 3 Opus — from AI safety practitioners and policy translators to animal-welfare technologists and career-switching students — with methodology details (logistic fit + k-means, holdout R², noise cutoff r<0.05) annotated.
Related event: Developer Clusters 1,711 EA Global Attendees Using Model Residual Streams(2 posts)→
More from Fun
- "If you come at the king, you better not miss": X users accused of photoshopping benchmark results — binarybits · 2026-10-25
- ChatGPT roasts user's 'stupidest' math intuition in viral screenshot — eigenrobot · 2026-10-25
- Ask ChatGPT to fill a photo of your empty bookshelf with books it thinks you'd own — venturetwins · 2026-10-25
- Creator jokes about agent-run life: showering, walking dog and booking flights all at once — ns123abc · 2026-10-25
- "Nvidia's H20 was traitorous" remark kills campfire chat — the stranger beside works at Nvidia — NathanpmYoung · 2026-10-25
- AI-enhanced FB Marketplace listings are deceiving buyers, warns Reddit user — bayls215 · 2026-10-25