GPT-3 can't quite pass a coding phone screen, but it's getting closer.
Conversation
Other questions are better for demonstrating its inhumanity - maybe I’ll put some of those together into a blog post
2
49
Show replies
Show replies
Impressive. Could you give some details about this chat setup and its priors? e.g. what makes it generate the intro/extro (e.g. "I am an AI created by..."). When does answer generation stop? Is there more context fed? Is it from a publication/cherry picked?
1
3
The prompt setup is everything up to “we’re going to do a phone screen”. It’s using the “davinci” model available in the openai beta. I didn’t really cherry pick too hard, this was just my last conversation before lunchtime.
1
7
It’s funny because I have been thinking how GPT-3 shows that Data’s struggles are not really the troubles that AI has. AI is great at creativity and emotion. It has trouble with common sense knowledge, being overconfident in its own knowledge, and staying on topic.
2
7
Show replies





