Re:OpenAI O3 breakthrough high score on ARC-AGI-PUB
by bubblyworld on Friday December 20, 2024 @02:23PM
(#42474053)
I think gaming the benchmarks is encouraged in the ARC AGI context. If you look at the public test cases you'll see they test a ton of pretty abstract concepts - space, colour, basic laws of physics like gravity/magnetism, movement, identity and lots of other stuff (highly recommend exploring them). Getting an AI to do well at all, regardless of whether it was gamed or not, is the whole challenge!
[ Parent
| on HN ]
There are no comments on this comment yet.