Decoder. plain-English AI glossary

Determinism

● Core

Same input, same output, every time — no surprises, no randomness.

Think of it like

A calculator: 2+2 is always 4, no mood, no dice.

Example

Set temperature to zero and a model becomes (mostly) deterministic — repeat the prompt, repeat the answer.

How it actually works

Deterministic decoding always picks the highest-probability token (greedy), removing run-to-run variation. It’s what you want for tests, caching, and reproducibility. True determinism can still be nudged by hardware and floating-point quirks, but temperature zero gets you most of the way.

For product teams

Reach for it when you need repeatable, testable outputs rather than creative variety.

For engineers

Greedy/argmax decoding yielding reproducible outputs; disturbed only by numerical nondeterminism.

Stochastic — The opposite is stochastic.
Greedy decoding — Achieved via greedy decoding.
Seed — Pinned with a seed.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome