cye131

3 ms·

1.
by cye131 1y ago
I'm not a fan of this blog post as it tries to pass off a method that's not accepted as a good or standard time series methodology (graph transformers) as though it were a norm. Transformers perform poorly on time series, and grap
2.
by cye131 1y ago
The new 2.5 Pro (05-06) definitely does not have any sort of meaningful 1 million context window, as many users have pointed out. It does not even remember to generate its reasoning block at 50k+ tokens. Their new pro model seemed to just t
3.
by cye131 1y ago
I'm curious whether anyone's actually using Claude code successfully. I tried it on release and found it negative value for tasks other than spinning up generic web projects. For existing codebases of even a moderate size, it burn
4.
by cye131 1y ago
These performance numbers look absolutely incredible. The MoE outperforms o1 with 3B active parameters? We're really getting close to the point where local models are good enough to handle practically every task that most people need t
5.
by cye131 1y ago
R especially dplyr/tidyverse is so underrated. Working in ML engineering, I see a lot of my coworkers suffering through pandas (or occasionally polars or even base Python without dataframes) to do basic analytics or debugging, it takes
6.
by cye131 2y ago
Does anyone actually have access to this? It says available for pro users on the website today - I have pro via my employer but see no "deep research" option in the message composer.
7.
by cye131 2y ago
Is it accurate to compare 8k example RL with 8k example SFT? RL with the same amount of examples would take massively more compute than the SFT version (though depending on how many rollouts they do per example). RL is more data-efficient b
8.
by cye131 2y ago
There is a surprisingly common use case for "quick and dirty univariate time series forecasts" that are basically equivalent to giving a small child a pencil, and asking them to draw out the trendline. The now-deprecated Prophet m