jumpCastle

It was a good benchmark until it entered the training set.

But open source like aider

Deepseek v3 was FP8

It's quite similar to muP

https://github.com/microsoft/mup

Stock prices cannot go up without if the planet is destroyed

Attention was invented because Bengio lab had to be disciplined about a black box (google had more compute)

Aren't services like runpod solve half of these concerns?

Are the updates really slow?

Spend more time with my side projects.

You can try his book. https://www.math.ias.edu/avi/book

You can use older models with the api.

Also the parameters are optimized also with loss of future tokens in the sequence.

Work pretty well for all classes of problems?

Rsu exist before ipo?

Use api, embed history and retrieve.

Use model output as training data. For better performance you can get some top log probs and minimize kl divergence.

Without open weights why would anyone care about them? By the time they could compete with gpt4 there's probably be gpt5 already.

The future is the AI also writes the high level descriptions of stuff.

But you can fine tune gpt 3.5 turbo, so your comparison is not clear.

I still use infinity for reddit though.

Gzip every query with all training data can get more expensive.

Silly is a charitable interpretation.

Title with a 10 digits number, meaningless first page figure and no experiments related to the main claim. Did a rogue author posted it without permission again?

They no longer require sign-in. Got scared of the reaction.

Is Azure as simple to use as OpenAI? Their documentation are much less simple.

For an answer I would expect it to get the same reward for both question orderings. So naively I would expect it to not be affected by the ordering.

Read codebase and implement feature. Read paper and prove conjecture.

https://github.com/google-research/long-range-arena

nat.dev

For web access there's nat.dev