SML — the smallest language model
Mar–May 2026
The question
Teacher → verifier → student
What we built
A large teacher writes candidate data; a verifier admits it; a small student is trained from scratch only on what was admitted. It is not distillation: nothing reaches the student that the verifier did not admit.
What it proved
TinkyBrain v6: 25.5 million parameters, 24 MB, about 2.9 s per answer on a 2017 Kindle Fire.
What it added to the brain
- Facts
Why it matters for defense
Only admitted knowledge reaches the student. That is the habit a defense system needs: nothing gets into the brain that a verifier did not let in.