Greedy Decoding: Always Pick the Highest-Probability Token, and Why It Fails

~10 min read

The simplest possible decoding strategy — always take the single most likely next token. Fast and deterministic, but it produces repetitive, low-quality text because it never considers the whole sequence, only the immediate next step.

Greedy Decoding: Always Pick the Highest-Probability Token, and Why It Fails is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Greedy decoding always picks the single highest-probability token at each step — deterministic and fast
  • It only optimizes the immediate next token, never the quality of the whole sequence
  • This myopic, no-lookahead behavior is exactly why it's prone to repetitive sentences and loops
  • There's zero randomness and zero recovery mechanism once the generation starts down a repetitive path
  • Best suited to latency-critical or narrow-answer tasks; rarely the right default for open-ended generation