Erdős problems
Systems closed several open problems that deterred strong mathematicians — the first new verifiable result rather than a restatement.
Not an assistant that finds papers and crunches tables, but an author: proposing the hypothesis, running the experiment and drawing a conclusion nobody knew before.
The moment AI advances science on its own is the inflection point of the whole timeline: progress speed stops depending on the number of people in the field.
Systems closed several open problems that deterred strong mathematicians — the first new verifiable result rather than a restatement.
Formal proofs give what the rest of science lacks: automatic verification, with no experiment and no trust in the author.
Autonomous labs supplied the hands; genuinely novel hypothesis generation is missing.
Moving the verifiable loop into chemistry and biology, where an experiment costs money and time rather than CPU seconds.
A peer-reviewed publication where the human contribution is only the choice of problem.