Another researcher says OpenAI trained on conversations, then claimed breakthrou

257 points · 15 comments on HN · read original →

Points and comments are a snapshot, not live.

Researcher accuses OpenAI of training on conversations then claiming a breakthrough.

A Bluesky post by timkellogg.me references a researcher accusing OpenAI of training on conversations and then claiming a breakthrough. The post links to a Mastodon thread by andreasthom. The article itself is a Bluesky post requiring JavaScript, displaying only the post text and metadata.

What commenters are saying

Commenters note the story is a screenshot chain: Bluesky of Twitter of Mastodon, lacking original links. Some joke about faking screenshots vs. video. Others point to duplicate HN threads and a moved discussion with original sources. A lone comment suggests models are simply good at short-scale math now.