What looks like “self awareness” is baked in during instruction tuning:
Basically turning any input into a document completion task, giving lots of examples where the completion contains phases like “I am an AI assistant”. This way, if GPT-3 would have completed your question with more questions that are similar, “assistants” will complete it with an answer, and one that sounds like it was spoken by someone who claims to be an AI assistant.
Basically turning any input into a document completion task, giving lots of examples where the completion contains phases like “I am an AI assistant”. This way, if GPT-3 would have completed your question with more questions that are similar, “assistants” will complete it with an answer, and one that sounds like it was spoken by someone who claims to be an AI assistant.