Residual-stream activations used to "ontologize" conference attendees, where SAEs fail
amplifiedamp · x · 2026-10-25
- WakaruLabs used a model's residual stream activations plus their "ontologizer" library to color-code and categorize EA Global conference attendees, to decide who to prioritize talking to.
- The author argues sparse autoencoders can't do this: SAEs yield a bag of 16k latents per person, while the ontologizer assigns each attendee a discrete code word, producing an interpretable typology instead of a feature dump.
More from Research
- arXiv paper: test collections may need expiration dates as neural rankers near human-level — bclavie · 2026-10-25
- FlashMap: flow map-based speculative decoding speeds up Qwen3-8B inference over 6x — caglarml · 2026-10-25
- CS papers near 18,000/month, up 10x since 2016, led by complexity and algorithms — burny_tech · 2026-10-25
- ICML Paper: Empowerment-Based Representations Are 1000x More Sensitive to Control-Relevant Features — burny_tech · 2026-10-25
- ETH Zurich teaches G1 robot to climb ladders with just 10 minutes of human data — Distinct-Question-16 · 2026-10-25
- New soft actuator weighs 15g, lifts 400g, costs under $3 to build — IanPritchard · 2026-10-25