Scientific originality

the scientific idea generation capabilities measured by our benchmark are poorly predicted by standard metrics of general intelligence scores.

Introduces LiveIdeaBench and finds scientific idea generation is largely decoupled from general intelligence — implying originality may need distinct training, sharpening the 'Einstein question'.

Some say we've entered a new age of AI-enabled scientific discovery. But human insight and creativity still can't be automated.

Reported survey directly on origination versus recombination: Hassabis says current systems 'can't do that'; Marcus calls much of the surrounding hype marketing.

AI (co-) scientists will only be truly effective when they can go beyond connecting words together, to modelling the full complexity of the systems those words describe.

A computing dean argues the ceiling on AI-originated science is that language-only systems don't model the underlying reality — surveying Robin, Co-Scientist, Sakana, and the fabricated-reference failure mode.