Another researcher says OpenAI trained on conversations, then claimed breakthrou

bsky.app

263 points·ColinWright··15 comments

Comments moved to https://news.ycombinator.com/item?id=49639408, which has the original source. I'll try to assemble the other links there too.

Also discussed here: https://news.ycombinator.com/item?id=49638353

Original posts

https://mathstodon.xyz/@andreasthom/117240535270608201

https://mathstodon.xyz/@andreasthom/117240536885387540

https://mathstodon.xyz/@andreasthom/117240537520615623

This story is someone on bluesky screenshotting someone on twitter without a link. The person on twitter screenshotted the source on the fediverse without a link. This is getting daft.

You can pretty accurately fake a video these days, faking a screenshot is 7 year old level of skills.

And here we are on HN discussing it.

Screenshotted this thread; will repost on Reddit and 9gag for the updoodles.

LOL Lets screenshot each others comments and comment on them.

Discussion is fine, let's just try actually linking back to the source.

A toot embedded in a tweet .

A toot in a tweet in a... skeet? Is there any consensus on what to call "a post on bluesky"?

Bleat?

Skeets is what the community wants, but the company does not (for obvious reasons).

flutter? idk

Here's the original Mastodon post:

https://mathstodon.xyz/@andreasthom/117240535270608201

[dupe] https://news.ycombinator.com/item?id=49638353

Nah the models are just that good at short scale math now. If your research program is close to it's goal the model can get it.

Reply on news.ycombinator.com