WingNews logo WingNews
top | new | best | ask | show | jobs
top | item 46490600

(no title)

bbatsell | 1 month ago

Anchor to new information (HN strips it from the URL): https://github.com/IQuestLab/IQuest-Coder-V1/issues/14#issue...

Context: Earlier this week a new model was released and researchers discovered that during training it had "cheated" on SWEBench by issuing git commands to find information it should have been blinded to.

Previous discussion: https://news.ycombinator.com/item?id=46472667

discuss

order

No comments yet.

powered by hn/api // news.ycombinator.com