Skip to content

Another researcher says OpenAI trained on conversations, then claimed breakthrou

bsky.app
261 pointsColinWright15 comments
On HN

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.

And here we are on HN discussing it.

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

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

LOL Lets screenshot each others comments and comment on them.

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

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"?

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

Bleat?

flutter? idk

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.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.