I talked to some experts about OpenAI’s new pitch for platforms to moderate content using GPT-4. They were surprisingly enthusiastic https://www.platformer.news/p/openai-wants-to-moderate-your-content
@caseynewton Makes sense. LLMs are bad at a lot of stuff, but this is language pattern matching, which is exactly the problem it's designed to solve.
@LouisIngenthron @caseynewton i think LLMs/AI can be a very useful tool… if they're used as a tool with human oversight and final judgment.
a weakness here will be that human adversarial language can adapt faster than language models.
@matunos @caseynewton Sure, but a strength is that the very platforms that need it most also happen to have a wealth of input training data that's up-to-the-minute on those adaptations.
@matunos @caseynewton Agreed, but that approach only requires a squad of humans, not a whole army.
@LouisIngenthron @caseynewton only insomuch as they are flagged as abusive… by humans.