Memory-assisted prompt editing to improve GPT-3 after deployment

TL;DR

This paper introduces a memory-assisted prompt editing method that enables large language models like GPT-3 to improve their accuracy through user feedback without retraining, by leveraging a growing case memory for error correction.

Contribution

It presents a novel approach combining memory and user feedback to enhance GPT-3's performance post-deployment without costly retraining.

Findings

Significant accuracy improvements on lexical and ethical reasoning tasks

Effective correction of misunderstandings through simulated user interactions

Demonstrates low-cost utility enhancement for large pre-trained language models

Abstract

Large LMs such as GPT-3 are powerful, but can commit mistakes that are obvious to humans. For example, GPT-3 would mistakenly interpret "What word is similar to good?" to mean a homophone, while the user intended a synonym. Our goal is to effectively correct such errors via user interactions with the system but without retraining, which will be prohibitively costly. We pair GPT-3 with a growing memory of recorded cases where the model misunderstood the user's intents, along with user feedback for clarification. Such a memory allows our system to produce enhanced prompts for any new query based on the user feedback for error correction on similar cases in the past. On four tasks (two lexical tasks, two advanced ethical reasoning tasks), we show how a (simulated) user can interactively teach a deployed GPT-3, substantially increasing its accuracy over the queries with different kinds of misunderstandings by the GPT-3. Our approach is a step towards the low-cost utility enhancement for very large pre-trained LMs. Code, data, and instructions to implement MEMPROMPT for a new task at https://www.memprompt.com/.