RL with Neural Feedback: training AI preferences purely from human brain signals
dr_cintas · x · 2026-10-30
Highlighted by drcintas as some of the craziest research this year: RL with Neural Feedback (RLNF) trains AI models to match human preferences purely from brain signals, no explicit ratings needed. Demo: a face image self-transforms in real time into a version the subject finds more attractive while they look at it — effectively swapping RLHF's reward source for live neural signals.
Related event: RLNF Trains AI on Human Preferences Using Brain Signals Alone(2 posts)→
More from Multimodal
- CapCut AI Extend workflow fixes cut-off ad endings: extend the close-up, sync the offer text — LearnWithBishal · 2026-10-30
- Claude Opus autonomously builds a walkable 3D IKEA room from product photos via HF agents — victormustar · 2026-10-30
- Pika H3 Flex generates 15s of 4K video in as little as 7 seconds — minchoi · 2026-10-30
- Free database of 1,407 AI girl outfits released, commercial use allowed — TinfoilTricorn · 2026-10-30
- AI cinema platform Fairground: viewers who pass the first 60 seconds watch 21 minutes — DavidmComfort · 2026-10-30
- kimo: a GUI-free video editor that AI agents drive over MCP, rendering frames to check their own work — LieTop7965 · 2026-10-30