> It's getting really hard to distinguish a GPT-3 text from actual human generated text.
No, it isn't.
What is getting really hard is to distinguish a GPT-3 text from actual human generated text without any context.
Plop GPT-3 down in a room full of humans having a conversation (or, under current conditions, put it in a Zoom call with humans having a conversation) and ask it to generate text as if it were another human participating in the conversation, and it will quickly become obvious that it is not a human, since it will fail miserably. The text it generates will be "human-like", but will have no relationship with what the humans are talking about.
I've actually done this experiment by putting a GPT-3 bot in a Telegram group. Its replies were mostly stuck in an uncanny valley where they sort of made sense, but often either lacked detail or seemed to very slightly misunderstand what the topic of discussion was. This might have been just because I didn't include enough context in the prompt, however. I have some plans for improving the prompting strategy, so we'll see.
I actually recently wrote a post on the topic of whether GPT-3 can be said to understand anything[1]. The argument is a bit too long to summarize here, but I don't think what GPT-3 is doing is as fundamentally different from what human brains do as people seem to think.
> I'm not sure "failed miserably" is a fair description either.
Interesting, thanks for the link!
Particularly towards the end, this looks similar to transcripts of conversations with ELIZA. I suppose "failed miserably" is indeed an overstatement, given that comparison, since ELIZA, IIRC, actually did fool some psychologists into thinking it was an actual paranoid human being.
No, it isn't.
What is getting really hard is to distinguish a GPT-3 text from actual human generated text without any context.
Plop GPT-3 down in a room full of humans having a conversation (or, under current conditions, put it in a Zoom call with humans having a conversation) and ask it to generate text as if it were another human participating in the conversation, and it will quickly become obvious that it is not a human, since it will fail miserably. The text it generates will be "human-like", but will have no relationship with what the humans are talking about.