Brian Lee
Brian Lee
Nothing actionable here, just recording some thoughts: Training from the last ~50 generations seems like an awfully long window. Early on in the training process, this seems like it would...
Instead of training on all game data indiscriminately, use the prior-Q and prior-policy recorded values to deduce how "unexpected" the actual played move was. If we filter for all the...
This will probably start as a private BQ dataset; will have to consult to figure out how to offer data publicly/how to allow the general public to query over a...
One of the symptoms of value net overfitting that we found was that the 8 board symmetries would yield wildly different results when put through the value net. So instead...
https://www.loom.com/share/eb6042214d1e45a1b6ca869586762722 probably because we mistake the null-ness for"not loaded yet" rather than "this is now null where it was previously not-null"
Some errors, like https://www.pantz.org/software/sqlite/unabletoopendbsqliteerror don't manifest until a user tries to label for the first time.
Dev console fills up with GET /tasks spam pretty quickly. Would be nice to drop poll frequency down to once/minute when no active tasks are known and then to 1/sec...
I really enjoy tossing in -u to my normal LLM prompts to get a good gut sense of cost. But `llm chat` doesn't support the -u flag. I _think_ the...