10 sats \ 7 replies \ @ek 29 Apr 2023 \ parent \ on: DarthCoin position regarding AI/chatGPT bitcoin
Just one example:
Showing it a piece of code and asking where the error is
And it shows you the error
Basically your personal Stack Overflow 24/7
Making effective use of chatgpt is very similar to making effective use of SO. The key to both is using them for your own personal parallel reconstruction.
Pasting a chatgpt response and saying “here’s what chatgpt says about X” is even lazier than linking to SO without providing context. It demonstrates laziness, and more importantly a lack of understanding of the topic.
Using either/both as tools to accelerate your knowledge on a topic, to the point you can confidently explain it in your own words, is simply making good use of the tooling available to us.
reply
You give up being creative and correct yourself or with other human help, into complacency and convenience of using AI.
That could not be considered built by you, the human, the developer, the coder. You are not a coder anymore, you are just a copy/pasta.
What if someday that AI is giving you deliberately a wrong answer ? And you take it as good... And that code will do harm later.
reply
Tell me you are not a coder without telling me you are not a coder, haha
reply
I see you added something to your meme:
That could not be considered built by you, the human, the developer, the coder. You are not a coder anymore, you are just a copy/pasta.
I didn't copy anything in that example. It just said "this could be the error" without giving me any code.
Then I fixed it myself.
What if someday that AI is giving you deliberately a wrong answer ? And you take it as good... And that code will do harm later.
In the context of code, it could deliberately give me a wrong answer, yes. But then I will notice because the code is not doing what I want.
To strengthen your point: Yes, it could include hidden bugs which I don't notice
Like one university once created malicious PRs into the Linux Kernel to test if the bugs will be spotted.
But so far, I only asked it about stuff which was very easy to verify since I then could look up that this is indeed the case how most people do stuff like this in this language which is new to me.
I see your point though, kind of forgot how easy it could just hide subtle bugs on purpose.
reply
I've found AI to be really helpful for generating boilerplate code or converting from one language to another. It's hit-or-miss for anything more complex than that.
You'd have to be a really bad developer to take what chatgpt spits out and just paste it into your IDE.. considering it probably won't even work straight away (this is the same for any info you pull off stackoverflow).
For AI to deliberately put malicious code in, it would have to be writing really complex stuff and not just code snippets, which as far as I can tell, it can't do.
reply
But then I will notice because the code is not doing what I want.
If you are asking in the first place the AI to fix your code, you will not be able to detect the deliberate error, you will just take it as good. Or you will not care anymore.
reply
You can always use it for writing code in new languages that you've never used before or that you are new.
Think of it as a spellchecker on steroids.
You still need to create the content, or at least the seed for the content, and then edit it to fit your goals.
And the beauty of code is that it's basically logic, so you can always run tests to ensure the code does exactly what it is supposed to do, and nothing else.
reply