By Gary Marcus
Publication Date: 2025-11-27 21:58:00
Latest news from famous machine learning researcher Ilya Sutskever:
Below is another summary of a just published article interview of him making waves, a little more technical. Basically, Sutskever says that scaling (achieving improvements in AI through more chips and more data) is leveling off and that we need new techniques; he is even open to it neurosymbolic Techniques and innateness. He clearly doesn’t predict a bright future for purely large language models.
Sutskever also said: “What I think is most fundamental is that these models somehow generalize dramatically worse than humans. And it’s super obvious. This seems like a very basic thing.”
Some of this will be news to much of the machine learning community; It might be surprising to speak of Sutskever, who is an icon of deep learning and who, among other things, contributed to the critical 2012 paper that showed how much GPUs can improve deep learning, the basis of LLMs, in practice. He…
