&Tutor: RL-Trained Tutor Whose Teaching Skills Emerge from Student-Learning Rewards Alone
alexisjross · x · 2026-10-31
humans& announces &Tutor, a tutor trained against their persimmon model. Key design: no pedagogical behaviors are trained directly — the model is rewarded solely on how much the student learns, so skills like scaffolding, pushing for rigor, and holding back all emerge implicitly from training.
The author frames it as a case study of the team's differentiated bet: having good models of users (learners) enables training AI that helps people over long time horizons, in an application close to their PhD roots.
Related event: Humans& Launches &Tutor: Model the Learner First, Then Train the AI Tutor(13 posts)→
More from Apps
- 60+ community-made ChatGPT slash-style words turn one word into a full visual style — heyshrutimishra · 2026-10-31
- ChatGPT tests chat resuming after app restart, screenshots show — dkundel · 2026-10-31
- A ready-to-play iOS TestFlight build of Torchlight II for phones surfaces via a random link — Dimillian · 2026-10-31
- Codex phone UI shown off: home-screen widgets track agent tasks and weekly usage limits — Dimillian · 2026-10-31
- Designer shows off new camera UI he created for the Gemini app — jessechand · 2026-10-31
- Set it on Sunday: a daily prompt that triages Gmail and briefs you via 45-second AI avatar video — toolstelegraph · 2026-10-31