BLAST RADIUS

Frontier Line and access · Benchmarks evals

Study Links Written Reasoning To Internal Patterns

Sep 12, 2026, first seen via The Decoder

A new study found that calculation, formula retrieval, and deduction correspond to distinct internal patterns in AI models.

Read the original at the-decoder.comOpens the publisher's site in a new tab

More in Frontier Line and access

Anthropic withholds latest model from UK testing agencySep 8Google DeepMind releases AlphaGenome AtlasSep 8OpenAI publishes proposed Navier–Stokes solutionSep 8OpenAI proposes mandatory safety rules for frontier developersSep 9Mathematicians Challenge OpenAI Mathematical MethodsSep 11