Short Answer

Contrasting Decoding Objectives

Imagine a text generation system has produced several potential output sentences. One decoding strategy selects the single sentence with the highest calculated probability. An alternative strategy selects the sentence that is, on average, most similar to all other high-probability sentences. Contrast these two strategies. Which one is more likely to avoid unusual or outlier generations, and why?

0

1

Updated 2025-10-10

Contributors are:

Who are from:

Tags

Ch.5 Inference - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Analysis in Bloom's Taxonomy

Cognitive Psychology

Psychology

Social Science

Empirical Science

Science