A demo of an AI model learning one person’s taste from their brain signals has passed 2.5 million views on X, and I think it deserves both the excitement and a few questions.
Rahul Chhabra’s company Sabi calls it RL with Neural Feedback. A wearable reads the viewer’s brain signals while they look at a generated face, the signal is used as the reward, and the face drifts toward what that person prefers, with no ratings, clicks or preference dataset. According to the thread, a simple embedding and nearest-neighbour model is enough, so no separate reward model has to be trained. The clip I saw shows a readout from disliked to liked as the face changes.
What I can’t tell from the thread is how many people were tested, whether the result holds up against simply asking people to rate the faces, or whether there is a paper yet. I also don’t know that a brain signal for arousal is the same thing as liking something, because a face can startle you as easily as please you. A stronger test would be several subjects and a blind comparison against ratings.
The part I find encouraging is that the company named the risks itself in the same thread: body dysmorphia, narrower beauty standards, bad actors steering feedback, and a plan for user consent and training on encrypted neural data. Personalization from brain signals could help people who can’t easily click or rate, and I would rather see it studied in the open with those questions answered.