ChatGPT seems amazing at first but it seems to be mixing in as much lies as truth. I ask it to generate code to do a thing and it looks correct at a glance but then I notice that it uses a function that does not exist. I ask it to fix the function because it uses a function that does not exist. It says that it was wrong and its going to generate the code again but correctly this time but it keeps failing again and again and making up shit. It ChatGPT a meme at this point?
It then claims the function is a private function (where? I cant find that function in any vaapi implementation or am I blind? or is it taking the function from proprietary private code?)
Ask it to write that function retard
Yes, it's only impressive at first glance. These LLMs are useless bullshit generators.
ChatGPT is a woman.
ChatGPT is not an artificial genius. ChatGPT quite literally does not possess the ability to think. You can ask it as much. ChatGPT is a very fancy autocomplete. It has been trained on a very large dataset of text and is able to infer how to program not from actually *knowing* how to program, or from *thinking* about how to solve a prompt, but from a statistical language model complete some text. It "knows" programming because software code was in its dataset. It "knows" syntax or libraries of Java or any other programming language because it has been trained on datasets that included code snippets and was able to identify patterns in text such as that, but it doesn't actually know anything
Wait, so you mean that it's not a person?
It doesn't seem to have a good language understanding either. It used vaCreateSurfaces incorrectly and then I asked it to fix it and it just rewrote the code in the exact same incorrect way
Because these AIs don't actually think so it doesn't know what is right and what is wrong, only what it thinks is a probable answer through brute force. It's really good at things that has lots of data to reinforce it and it falls apart if you pull the string at the edges. You will always have to be extra diligent with these AIs because they produce plausible results but like with the AI images, if you start looking closely things don't make sense.
which is weird, because it was unable to see that the types it calls a functions with doesn't match. But it is able to know the correct data to pass to a void pointer data at index position 3 because that was mentioned in one way in a documentation somewhere, which it was able to find because it looked at the function declaration
reminder that you are beta testing for them to help learn their models. Eventually they will cut it off and charge companies for their data that you helped train 🙂
I think that seems likely
for all we know they added a rule to chatGPT to say that just like how they added a rule for it to filter offensive content
>ChatGPT is designed to have the ability to generate code in various programming languages. The accuracy of the code generation depends on the specific task and the quality of the input provided to ChatGPT. ChatGPT is able to generate code by utilizing its understanding of programming languages and its ability to process and analyze natural language input. It can then generate code that is syntactically and semantically correct, and optimized for the specific task at hand. However, it is important to note that ChatGPT's code generation should be thoroughly tested and reviewed before being used in any critical systems.
>It ChatGPT a meme at this point?
It's publicly available. What do you think we are capable of that isn't publicly available?
To be honest, I think that ChatGPT is still better than google search (or any other search engine) when it comes to finding information related to how to use different libraries and APIs together. Google search is terrible when searching for anything low level.
It does this for me too. I gave up trying to use LUA because it wasn't actually learning anything I was teaching it, even after providing it with the error code it will still give you the same thing just using a different method of getting that error.
It works well for certain things.
I got it to generate a hilarious short story earlier.
What you can do is to ask it to play a game where you complete each other's sentence.
It has been known for a long time that to dupe normies you just need to be confident, not right. ChatGPT is very confidently wrong
It tells you the truth whenever it tells you that it's just a large language model.