ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2306.07567
20
0

Large Language Models Sometimes Generate Purely Negatively-Reinforced Text

13 June 2023
Fabien Roger
    SILM
ArXivPDFHTML
Abstract

When using adversarial training, it is common practice to train against the most egregious failures. However, this might imply using examples with sensitive information (such as leaked passwords or security vulnerabilities) as training data. One might assume that language models trained with gradient descent never generate text snippets which were only present in examples associated with the lowest possible reward. In this paper, we show that this assumption is wrong: in some situations, large language models do learn from such negatively-reinforced examples. We present a specific training setup that enables Pythia-160M to guess passwords 13% more often than it would by guessing randomly, despite only showing it these passwords on examples where the model is incentivized to not output these passwords. Our code is available at www.github.com/FabienRoger/Learning-From-Negative-Examples

View on arXiv
Comments on this paper