from Hacker News

"Devin" AI automates Upwork job, making inferences on a computer vision model

by drubio on 3/12/24, 10:41 PM with 1 comments

  • by ben_w on 3/12/24, 11:47 PM

    The SWE-bench graph (Devin: 13.85%, Clause 2: 4.8%, GPT-4: 1.74%) looks surprising, perhaps even a bit suspicious (too far above the next best) — can someone elaborate further?

    Perhaps it really is this good, perhaps it's just my anxiety looking for reason to doubt. It was only 9 hours ago I was writing:

    > It's certainly still possible today that some random individual has a crucial insight that gives them an edge over the big names

    - https://news.ycombinator.com/item?id=39679669