What is determinism in AI?

Determinism means the same specified result follows from the same input and relevant conditions. An AI claim should identify the component, result and fixed conditions it covers.

Updated October 11, 2026

What does “the same conditions” mean?

For this reference, it includes relevant state and versions, not just a prompt’s words. PyTorch documents that complete reproducibility is not guaranteed across releases and platforms or between CPU and GPU, even with matching seeds. Source: PyTorch.

Thinking Machines Lab also distinguishes temperature-zero sampling from the reproducibility of practical LLM inference. Read the introduction.

Is determinism the same as correctness?

No. As an illustrative example, a calculation that consistently adds an incorrect fee is repeatable but wrong. Use separate checks for the required answer, repeated execution and records of what happened.

Read the full deterministic AI definition, LLM determinism guide and testing protocol.

Reading scope

• PyTorch: Introduction and initial reproducibility guidance read; main documentation can change. Accessed 2026-10-11.
• Thinking Machines Lab: Introduction read; cited for the distinction between temperature-zero sampling and reproducible inference.