BDH-CQ — a 150M-parameter system that learns a puzzle rule from the worked examples in front of it and does its thinking as numbers, never as words

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

1. 1BDH-CQ0:00
2. 2The worked examples are the only place the rule exists0:13
3. 3The model that finishes a puzzle is identical to the one that started it0:39
4. 4The examples go into one memory that does not grow0:58
5. 5Only the last pass becomes an answer1:28
6. 6BDH-CQ solves 29.5% of 400 public puzzles for a fraction of a cent2:13
7. 7One more worked example takes sorting from 0 of 24 to 13 of 242:43
8. 8The reasoning stays as numbers inside the model until the last pass turns it into an answer3:15