Short Answer

Limitations of Minimal Data Fine-Tuning

A research team successfully fine-tuned a large pre-trained model to follow instructions for summarizing news articles using only 100 high-quality examples. However, when they later asked the same model to write a short poem, it produced a summary of the word 'poem' instead. Based on the principles of data-efficient fine-tuning, explain the most probable reason for this specific failure.

0

1

Updated 2025-10-10

Contributors are:

Who are from:

Tags

Ch.4 Alignment - 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