jumpCastle

6 ms·

1.
by jumpCastle 1y ago
It was a good benchmark until it entered the training set.
2.
by jumpCastle 1y ago
But open source like aider
3.
by jumpCastle 1y ago
Deepseek v3 was FP8
4.
by jumpCastle 1y ago
It's quite similar to muP https://github.com/microsoft/mup
5.
by jumpCastle 2y ago
Stock prices cannot go up without if the planet is destroyed
6.
by jumpCastle 2y ago
Attention was invented because Bengio lab had to be disciplined about a black box (google had more compute)
7.
by jumpCastle 2y ago
Aren't services like runpod solve half of these concerns?
8.
by jumpCastle 2y ago
Are the updates really slow?
9.
by jumpCastle 2y ago
Spend more time with my side projects.
10.
by jumpCastle 2y ago
You can try his book. https://www.math.ias.edu/avi/book
11.
by jumpCastle 2y ago
You can use older models with the api.
12.
by jumpCastle 2y ago
Also the parameters are optimized also with loss of future tokens in the sequence.
13.
by jumpCastle 3y ago
Work pretty well for all classes of problems?
14.
by jumpCastle 3y ago
Rsu exist before ipo?
15.
by jumpCastle 3y ago
Use api, embed history and retrieve.
16.
by jumpCastle 3y ago
Use model output as training data. For better performance you can get some top log probs and minimize kl divergence.
17.
by jumpCastle 3y ago
Without open weights why would anyone care about them? By the time they could compete with gpt4 there's probably be gpt5 already.
18.
by jumpCastle 3y ago
The future is the AI also writes the high level descriptions of stuff.
19.
by jumpCastle 3y ago
But you can fine tune gpt 3.5 turbo, so your comparison is not clear.
20.
by jumpCastle 3y ago
I still use infinity for reddit though.
21.
by jumpCastle 3y ago
Gzip every query with all training data can get more expensive.
22.
by jumpCastle 3y ago
Silly is a charitable interpretation.
23.
by jumpCastle 3y ago
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?
24.
by jumpCastle 3y ago
They no longer require sign-in. Got scared of the reaction.
25.
by jumpCastle 3y ago
Is Azure as simple to use as OpenAI? Their documentation are much less simple.
26.
by jumpCastle 3y ago
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.
27.
by jumpCastle 3y ago
Read codebase and implement feature. Read paper and prove conjecture.
28.
by jumpCastle 3y ago
https://github.com/google-research/long-range-arena
29.
by jumpCastle 3y ago
nat.dev
30.
by jumpCastle 3y ago
For web access there's nat.dev

More ›