My original tweet makes the performance look a bit better than it is, if you're looking for exact answers. Here's the accuracy in terms of exact solutions.

Mar 14, 2023 · 1:19 AM UTC

Replying to @colin_fraser
The accuracy for d1>d2 division increasing with output length is very interesting! Could you share some samples of the numbers being used here, labelled by success (were successful divisions with “easy” d2s like 1 or 2 or 5 or 10)?
Replying to @colin_fraser
What’s an example of an unparseable response?