Greedy decoding
Always picking the single most likely next token. Fast and repeatable, but prone to dull or looping text.
Also: beam search, argmax decoding
Greedy decoding takes the top token at every step, equivalent to sampling at temperature zero. It is deterministic in principle and suits tasks with one right answer, such as extracting a date. For open-ended writing it tends to repeat itself. Beam search keeps several candidate continuations in parallel and picks the one with the highest overall probability; it was standard in machine translation but is less used for chat. In practice, even greedy outputs from hosted APIs can vary slightly because of how calculations are batched on servers.