Goalposts

Hacker News has set AI a lot of challenges over the years. Which ones has it met?

2025 AugustCamperBob2

About how to define AGI, given there is no consensus definition.

A common test is something like ARC-AGI 2, where even the best models struggle to get into double-digit percentages. One thing I'd expect from AGI would be to go beyond "Solve this ARC-AGI test" to "Come up with more tests like ARC-AGI."

It can't have been lost on the humans who worked to create those tests that their own jobs should be automatable.

An AI comes up with new tests like ARC-AGI, not just solves them.

Has this happened?

Yes 17 (13%)Not sure 61 (46%)No 54 (41%)

Votes cast 1–2 October 2026: 100,590 votes from 9,694 people.