News for hackers. Stuff that matters.
Re:ARC-AGI Leaderboard by johnfn on Saturday July 25, 2026 @12:18PM (#49048884)
How do you propose they check for something like this? They can't exactly ctrl-f the model weights for "Arc-AGI".
Re:ARC-AGI Leaderboard by vickychijwani on Saturday July 25, 2026 @02:53PM (#49050455)
Anthropic expends tons of compute and effort on understanding internal model states [1]; this kind of thing is right up their alley.

[1]: Recent example: https://www.anthropic.com/research/global-workspace

Re:ARC-AGI Leaderboard by phoghed on Saturday July 25, 2026 @12:56PM (#49049252)
> they can’t possibly know or find out what was in the training data

doesn’t appear to be a very strong argument