English

The Scaling Properties of Implicit Deductive Reasoning in Transformers

Artificial Intelligence 2026-05-07 v1 Computational Complexity Logic in Computer Science Symbolic Computation

Abstract

We investigate the scaling properties of implicit deductive reasoning over Horn clauses in depth-bounded Transformers. By systematically decorrelating provability from spurious features and enforcing algorithmic alignment, we find that in sufficiently deep models with a bidirectional prefix mask, implicit reasoning approaches explicit CoT performance across graph topologies and problem widths, though CoT remains necessary for depth extrapolation.

Keywords

Cite

@article{arxiv.2605.04330,
  title  = {The Scaling Properties of Implicit Deductive Reasoning in Transformers},
  author = {Enrico Vompa and Tanel Tammet},
  journal= {arXiv preprint arXiv:2605.04330},
  year   = {2026}
}

Comments

preprint

R2 v1 2026-07-01T12:51:54.761Z