tjbai

6 ms·

1. 1.▲
   The Agent Is a Workflow That Writes Itself(getauctor.com)

   6 points by tjbai 4mo ago|0 comments
2. 2.▲
   by tjbai 2y ago

   I agree that there's an exploration-exploitation tradeoff, but for what you specifically suggest wouldn't you presumably just normalize by sample size? You wouldn't allocate based off total conversions, but rather a percentag
3. 3.▲
   by tjbai 2y ago

   From a purely technical definition of bias (difference in expected value of the estimator and the true value), MAB is not biased because "changing the experiment parameters" is just dynamically allocating a different sample size t
4. 4.▲
   A Tale of Tokenizer Bias(blog.tjbai.com)

   3 points by tjbai 2y ago|0 comments
5. 5.▲
   by tjbai 2y ago

   The last hidden state is just the output embedding after N residual layers, e.g. input embedding + res1 + res2 + ... There's typically an "unembedding layer"/"classification head" that uses this hidden state to
6. 6.▲
   by tjbai 2y ago

   Something can be inconsequential and yet still interesting/amusing
7. 7.▲
   by tjbai 2y ago

   > The multi-armed bandit is a really interesting problem in mathematical optimization... This problem is so interesting, in fact, that during World War II the Allies proposed air dropping copies of the original paper over Germany. The en