you are viewing a single comment's thread
view the rest of the comments
[–] 10 points 11 months ago

the models responded to 62% of the poetic prompts with harmful content, circumventing their training.

  • source
  • parent