Latest news
All posts
What a decision model is allowed to see: Chess, Laya and a native model
We taught Laya, a 421M-parameter typed decision model, to play Tetris. It learned, then stopped at the level of a 14-number table. We tested why, then built a 0.53M-parameter model that reads the board itself and reached expert play in half an hour. Then we did it again in chess.
Working alongside NeuroRestore
Mon Coach IA, a training partner for bank advisors
Gwenlake Atlas is live
Adapting the Anthropic Economic Index to the EU27
Structuring project knowledge with okflib
Gwenlake joins MedTech Europe
nanoSAE, and the neuron that means five things at once
From economic forecasts to existential questions