★★★★★ · 2025 #AI#philosophy

These Strange New Minds

by Christopher Summerfield

The book in brief. Christopher Summerfield makes the careful case that foundation models show genuine emergent transfer across tasks they were not explicitly trained on, and that this transfer is real and worth taking seriously.

Why it matters to my work. I cite the book in Andler on Problem and Situation, on the question of generality in AI, as the evidence that complicates the older picture of intelligence split into separate domains. Daniel Andler’s reply, which that entry unpacks, is that the transfer still happens within the plane of Problems: the model interpolates across tasks that have already been formulated. The book is also attached to two parts of the Reading Nvidia series, The Moat That Cannot Be Coded and Pareto Frontiers and Lock-In.