Testing.
https://join-lemmy.org/docs/users/03-votes-and-ranking.html
Edit: I was wrong the ranking that works like forums is New Comments and yes it seems to take into account the OP comments.
The paper actually demonstrates a 16-million context window with 92% accuracy. Most models can be retrained to have a 100k context window with over 92% accuracy, but the accuracy drops to 74% at 256k. The code has already been released on GitHub as well. I'm excited to see the development of 100k models using this method soon!
Comparing current LLMs with autocomplete is stupid. An autocomplete can't pass law or biology exams in the 90th percentile like GTP-4 can.
Hopefully there are some people more positive than that, willing to change society so AGI doesn't make most humans starve to death or be imprisoned.
I feel like this is what happened when you’d see posts with hundreds / thousands of upvotes but had only 20-ish comments.
Nah it's the same here in Lemmy. It's because the algorithm only accounts for votes and not for user engagement.
I personally prefer it colorful as it is, or even more colorful like Matt Wolfe's Midjourney generated thumbnails. It's a nice change of pace from the usual thumbnails.
@InternetPirate
@lemmy.fmhy.ml