Hacker News Reader

Theyre saying if you find references to a very specific set of phrases that were probably included accidentally on github then github is likely part of the training data.

GitHub is obviously part of the training data, you don't need to find obscure tokens to tell.

And it's nothing new.

https://github.com/jiangyy/gpt-tokens

People found these adult-site-related Chinese phrases in Gpt-4o. The OP is more than one year late.

FWIW, I didn't get that sense.

Wouldn't it be best for them to strip that out of the training data for moderation reasons?