Scaling-law critique argues data-driven ML cannot reach symbolic reasoning alone

A new cs.AI paper argues that data-driven machine learning has limits when it comes to symbolic-level logical reasoning. The authors frame the claim as a challenge to simple scaling-law optimism.

The argument is relevant because frontier AI progress is often described as a function of more compute, more data, and larger models. If some reasoning capabilities require different structures, evaluation and architecture choices need to reflect that.

The paper adds to the long-running debate over whether neural scaling alone can deliver robust abstraction and logic.