Comment on Announcing ARC-AGI-3 - An benchmark that tests if AI can explore, learn, and adapt in unfamiliar situations. Humans score 100%. Frontier AI scores 0.26%.
Stress test it. Low, average, high, impairment conditions, safeguards off, order, chaos and everything in between.
I haven’t read all of their Benchmark introduction and Technical Documentation. I assume you have and didn’t find any of the tests you’re asking for?
gnufuu@infosec.pub 10 hours ago
I haven’t read all of their Benchmark introduction and Technical Documentation. I assume you have and didn’t find any of the tests you’re asking for?