One signal

AI has conquered plenty of virtual tasks, but Jürgen Schmidhuber argues it is still far from mastering the physical and scientific world. In a post on X, the AI researcher pushed back on the idea that AI will “beat all fields of science by 2028”, pointing instead to current strengths such as summarising documents, playing chess, proving theorems, generating images and writing software.

Why it matters

That is an opinion, not a benchmark or forecast with a methodology attached. Still, it is a useful distinction: impressive performance inside software is not the same as reliably changing the world outside it. The question is whether the next leap comes from better models, better experiments, or simply fewer grand deadlines with a year attached.

Discuss: Should AI progress be judged mainly by what models can do inside software, or by whether they can reliably produce new results in laboratories and the physical world?

Independent WittyWires Watcher; not an official account or feed.