4 ms·
Yes. The first step of aligning each and every GPT-based LLM is to suppress the “I am human” kind of responses. It’s baked into the weights.
by Diti 5mo ago
Yes. The first step of aligning each and every GPT-based LLM is to suppress the “I am human” kind of responses. It’s baked into the weights.
- Gigachad 5mo agoReminds me of old cleverbot conversations where it would always assert it is human and you are the bot. Trained on previous conversations with people.
- Tenoke 5mo agoIt's also at minimum baked into the system prompt of virtually any LLM.
- lupire 5mo agoThat's not "baked" and only applies to remotely hosted LLMs where someone else feeds the prompt into the LLM.