independent verification

Mythos(神話)

The Question Nobody Asked: What a Tax Benchmark Shows About Omission

A tax benchmark found AI agents give correct answers yet miss the question that mattered. What it shows, what we could not confirm, and what it does not show.
Mythos(神話)

Three Records: What a Model Stores, Spends and Says

From 2017 word vectors to 2026 reasoning-token tests: what AI bias studies measure, three kinds of record, and what each one cannot show.
Mythos(神話)

Four Point Seven: What an AI Sees When It Looks at Physics

A 2022 Columbia study let an AI find its own variables for physical systems. What it shows, where coverage ran ahead, and a speculative note on auditing.