Hacker News new | ask | show | jobs
by dclowd9901 1048 days ago
Is it not the case that AI is pretty good at detecting AI generated content?
5 comments

OpenAI does not think so.

> As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy.

https://openai.com/blog/new-ai-classifier-for-indicating-ai-...

It is not, which is why lots of people have been focusing on watermarking–which, obviously, all kinds of actors that don’t feel the need to comply with regulations, either because they are exempt (practically, even if not theoretically) governments, or “renegade” anti-establishment will simply either not implement or not provide the public with the pattern information needed to detect the watermark, so...
If so, it's only for one generation, until they add "tricks current AI detection" as a metric for the next training.
It's not. One of the things you can train an AI to do is to generate output that isn't recognized by a given "AI detector."
It is not the case.