> You realize that near human-level AI for $20/month is a bargain
ChatGPT isn't remotely “near human-level AI”, though.
ChatGPT isn't remotely “near human-level AI”, though.
Yes.
> It's pretty close.
No, its not, even in its narrow domain, whose narrowness by itself demonstrates its not anything like a near-human-level intelligence.
> Its not always correct but neither are humans.
Its a very good text generator, and is very useful. Its nothing even similar to a human-level intelligence.
In short, the code it produces has to be reviewed by programmers who know what they're doing. Sometimes that speeds things up to have it produce the code and we review it, but sometimes it slows things down and it's easier to just write it yourself.
I see ChatGPT as an extremely powerful tool that can boost a programmer's productivity, but despite its incredible capability it cannot be trusted in ways that we can trust human programmers.
In fact, I'd argue it might already be superhuman in the same way the Ocean of Solaris could be. Think about it. The Ocean can't drive or do arithmetic either. It is never shown that it evolves over time.