Gareth Davidson

So.. I pulled this and installed it locally, there's loads of juicy stuff in there. I'll see how viable it is to snag some Q+A data sets out of it...

If you feel a prompt is too complex or it feels a bit like asymmetric warfare by the prompter, then simply skip it. If enough people skip a prompt it'll...

I think releasing the data in full, with only legal things like PII removed, and with users and votes identified by hash or something would help. Then we have something...

I've updated with some feedback. I'll actually do a smallish one and iron out the process before mass-editing people's data issues - might take a few days so more feedback...

> New text seems a little marketing speak too and buzzy imo. I agree. When it comes to writing docs I tend to think "it's not done when there's nothing...

I don't think the *assisted* suicide thing is that big of a risk or something we should block. If someone uses the thing to research their options, get referred to...

If you take NSFW content from the web, edit it and then train models to think that's CSAM, then wouldn't that be completely out of distribution? And wouldn't models trained...

I'll repeat what I said on Discord here: The risk with editing people's words and making them into paedophile content, then sharing it as a CSAM data set, is that...

> This is something I'm going to try out and see. @shahules786 can give more info about this. Even if the original text doesn't flag as CSAM in our model,...

> If we build a system to flag CSAM, which metric should it be more careful about, false positives or false negatives? If you were to pick one to optimise...