Hacker News new | past | comments | ask | show | jobs | submit login
Tried to replicate Chinchilla paper and discovered discrepancies (twitter.com/tamaybes)
12 points by tosh 10 months ago | hide | past | favorite | 4 comments



I only can see the first post, which is basically useless. I see its a 1/9 post but Twitter is no longer a good place to have these kinds of discussions.

Nitter.net is also dead, so my #1 way of reading Twitter posts is gone. I've never posted to Twitter before and rarely read it, so I don't expect to ever need an account.


You can find the preprint, which the Twitter thread summarizes, here (https://arxiv.org/abs/2404.10102)


One of the replies to the thread (that you also can't see signed out) contains a link to an "unrolling" of the thread:

https://threadreaderapp.com/thread/1780639257389904013.html?...


Since they extracted the data shown in a chart and found it didn't match the equations, I'm wondering if this might just mean the chart is messed up.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: