Discussion sur la contournement des filtres NSFW de ChatGPT

From a technical standpoint, the filters are integrated deeply in the model's response generation. Bypassing them is not straightforward and may require modifying the underlying model, which is not accessible to most users.

Isn't it better to focus on improving the AI to understand context better rather than trying to bypass safety measures?

I think these filters are in place to ensure safe and responsible AI use. Trying to bypass them might lead to violations of OpenAI's policies.

I heard about users trying to use code or prompt engineering to trick the filters, but that often results in unstable or unpredictable outputs.