On: OpenAI O3 breakthrough high score on ARC-AGI-PUB

[Terr_]

Why should one kind of phenomenon which slows down performance on the test be given a special "you're more intelligent than you seem" exception, but not others?

If we are required to break the seal on the black-box and investigate the exactly how the agent is operating in order to judge its "intelligence"... Doesn't that kinda ruin the up-thread stuff about judging with equations?

[Earw0rm → Terr_]

The point is that the wall-clock measuring the test-taker's performance should be in the same context as the test-taker.

They will still complete the task in 70 local minutes, even if that's eighty of an outside observer's.