NVIDIA AI Account Shares AVO ARC-AGI-3 Results
Official NVIDIA AI account reports its coding agent completed the benchmark without instructions or rules.
TLDR
NVIDIA's official AI account posted that its general-purpose coding agent AVO scored 100% on the ARC-AGI-3 interactive reasoning benchmark. The post states the agent completed all 183 levels across all 25 public environments by figuring out actions on its own. No instructions, explicit rules, or stated goals were provided. The announcement includes a photo. User @scaling01 replied that even Nvidia is participating and linked to the post.
Combined views
895.9K
3 Sources, first seen ago
Sources
LA 
Lisan al Gaib@scaling01Even Nvidia is participating https://twitter.com/NVIDIAAI/status/2090786258981466231
- likes: 540
- replies: 6
- bookmarks: 48
- reposts: 13
CS Christian Szegedy@ChrSzegedyThe next benchmark should be on self-esteem. https://twitter.com/NVIDIAAI/status/2090786258981466231
- likes: 22
- replies: 0
- bookmarks: 4
- reposts: 0
BB bbuerk@bbuerk1@NVIDIAAI Scoring 100% on a public dataset is meaningless. Not saying that the agent couldn’t do the same on the private dataset, just that I won’t care until I see it
// //@fuckithonestly_@NVIDIAAI Can't wait for the French faggot to be like umm actually not like that teehee
MS Manish Sainani 🤫@manishsainani@NVIDIAAI Yes 🙌 congrats 🍾
SH shri@lazybanananann@NVIDIAAI time to make ARC-AGI-4