Flappy bird is more complicated than pong, and GPT4 apparently did it in 'one shot'. I think there's improvement being made, even if it's not as impressive as you'd like it to be.
It's really not impressive to me. Humans are gullible and want to anthropomorphize computers. Every time someone claims a LLM like GPT-* can do something apparently amazing, we're seeing the ELIZA effect (https://en.wikipedia.org/wiki/ELIZA_effect) in action.
Of course, it's not an apples-to-apples comparison. But I don't have access to GPT4, so I can't check how it does on Pong written in PyGame.
So the more versions of Flappy Bird are online, the easier for ChatGPT to generate something that works. Difficulty of problem doesn't matter.
I'm amazed how much resistance there is to one of the biggest technological advancements in history. And this is JUST getting started.
I think the answer to this question is yes? If the developer can find a working example that they can copy paste (which is what GP is saying GPT-4 is essentially doing).
It’s actually just as likely for GPT-4 to have pasted a broken code example than a working one; it doesn’t understand if the code is correct.
Sounds familiar