AI Lab
This lab hosts ongoing experiments in applied artificial intelligence for clinical education. These are working tools — built, deployed, and maintained by a clinician — not proofs of concept.
Virtual Teaching Assistant
A pilot teaching assistant designed for seminars on dream interpretation. Optimized for discussion, synthesis, and learning support. Built on large language models, it offers an interactive seminar mode where learners can ask questions, request syntheses, and explore alternative readings in a conversational format.
The assistant is scoped for education and demonstration — not clinical decisions, diagnosis, or patient care.
Launch Teaching Assistant → Login required
Nanochat Lab
A small language model sandbox built from Andrej Karpathy’s Nanochat tutorial codebase, designed for “knob-turning” demonstrations. The model was pretrained on a large sharded text corpus, with additional synthetic identity chats and instruction tuning.
Stable
Baseline decoding defaults. Use this as your reference point for coherence and consistency.
Open Stable →Unstable
Same checkpoint, different decoding defaults. Try the presets — Stable-ish, Spicy, Chaos — to see how temperature, top-k, and max tokens trade off coherence versus novelty.
Open Unstable →How to use the lab
Start with Stable as your baseline. Then open Unstable and try the presets to see how decoding parameters shift tone, consistency, and drift. The best demonstrations come from repeating the same prompt across settings and comparing the results.
Temperature, top-k, and max tokens are visible on every page. On Unstable, you can adjust them live and watch behavior change in real time.
Need access? Contact Oscar for the guest login.