CVPR 2026 ReLearn workshop: Efros argues AI may already learn too much from humans
TheZachMueller · x · 2026-10-08
At the CVPR 2026 ReLearn workshop co-organized by Lambda, researchers including Alexei Efros, Dima Damen, Manling Li, Alan Yuille, Saining Xie and Bill Freeman debated what machines should still learn from human intelligence.
Key points:
- Efros: modern AI may already learn too much from us — foundation models are "cultural technologies" distilling human knowledge, and even self-supervised vision relies on human-designed augmentations.
- Damen: advocated egocentric vision — learning from the continuous experience of a single life rather than static datasets.
- Li: mapped where foundation models still fail at spatial understanding, from topological to metric reasoning.
The overall arc: the field is shifting from "what to learn from humans" to "how machines should represent and experience the physical world."
More from Research
- Model Spotted Reusing Its Own Prior Results in Reasoning Traces, a Spark of Theory Building — QuintinPope5 · 2026-10-08
- RLHND turns video diffusion models into physically grounded hand trackers for robot learning — RLWRLD · 2026-10-08
- Paper models LLMs as oracles in a pushdown automaton, unifying Agents and Workflows — China666 · 2026-10-08
- OpenAI model proves Hilbert's Tenth Problem false over Q, sidestepping 80-year approach — aran_nayebi · 2026-10-08
- LLMs reproduce classic rat experiment, seeking steered-positive zones and shutting off bad states — awjuliani · 2026-10-08
- NEEDLEwork stitches suboptimal robot demos into better training data — SongShuran · 2026-10-08