Where it stands
SAIL — the Socratic AI Learning Lab — was proposed as a €3M frontier-lab bid to SPRIND's Next Frontier AI Challenge in May 2026. It didn't advance, and neither did two follow-on applications. This site isn't the pitch deck for that bid. It's the research thread that survived it, still moving, without the funding.
Sailing, not a finished vessel — a small, community-driven experiment, in progress, not a funded lab with a building and a staff.
The bet: SAITO
The Socratic AI Training Objective reformulates what an AI system optimizes for. Not immediate task completion, not rated helpfulness — measurable capability gain in the human it's working with, checked weeks later, not in the moment.
Three coupled questions: what to ask (productive difficulty, not just the fastest answer), who you're asking (a learner's actual state, not an average), and when to ask it (spacing and retrieval practice work over weeks, not within one exchange).
What's real right now
No claims here that aren't running somewhere.
A Socratic scaffold agent, live in a real chat community Real content-aware judgment, not scripted heuristics Local-model generation, Thompson sampling over question styles A Julia/Pluto notebook teaching the mechanism from real data
The next real site of work is renate.chat, a vocational-training assistant already carrying real trainer questions and an eval-gated deploy discipline — SAITO's eventual pilot, not yet its home.