An independent notebook on machine intelligence Vol. I · Nov 2026

Careful measurements of uncareful machines.

Tensor Factory is a small, independent bench for testing AI systems: where they look, what they invent, and what they refuse. Methods are published, code is open, and results are dated — because everything here will be wrong eventually.

No. 002 1 min read

Confidently wrong: how often do language models make things up?

A hallucination benchmark across eight frontier models, 1,200 questions, and one uncomfortable chart.

The question isn't whether models hallucinate—they all do—but how the failure is distributed. Some models fabricate rarely but grandly; others hedge on everything.

Fig. 1 — Fabrication rate across 8 evaluated models. Lower is better.

Continue reading →