Great quote from the The Vergecast about AI (in a product review of an AI device that queries other AIs as needed):
David Piece explains that after getting an answer from the device he has to get his phone and google the results to make sure it is accurate - "It completely defeats the purpose. We are so far away from these things being reliably honest, and reliably fast, and reliably useful that it is like 'what's the point?'"
You're giving it very constrained inputs and can easily test the output. Is the bug gone?
I'm no software engineer but I've spoken with a couple of people who have found it useful to track down stuff a lot faster than they can do it on their own.
@sapphire@jwcph I heard from a friend that was very bullish on AI originally that he found that difficult. He does some Arduino programing, and he could tell it to change x in the program, but then would have to re-proofread the entire thing because it might change some other bit of code that was already correct. It was like it was rerunning the prompt with the added comment, instead of freezing the output that he had no problems with.
@sapphire@Urban_Hermit Again: Pretty factual. Did it report back the correct lines, and nothing else? Did it obey you when you told it to change nothing? As soon as you're asking something that the thing could "hallucinate" - i.e. lie - about, you're going to have to re-check everything.
Please can you help me understand, as a non-coder? I know html. If I have a text file with 1000 lines and ask AI to find the bug and tell me what lines it changed, could I not then just copy those specific lines from its output into my text file and test the page? Then I could see if the change was made as desired, and know the remaining html stayed untouched, since I only pasted specific lines? And if it's not a good fix, it'll be obvious quickly?
It's usually relatively easy to locate the kinds of bugs that AI can detect using existing tools. If something isn't compiling or there's a memory error, existing tools tend to be able to give you the red wavy underline or point at the missing semi colon.
If my code is producing an undesirable output, that typically means that I am not specifying what I want my code to do in a way that the computer understands. If I can't say what I want it to do to python, I'm not going to be able to explain it to a machine.
Reading the docs as I go through the code to identify what misunderstanding I had has utility with a pretty quick pay off.
I won't feel awful or open myself up to legal liability related to Microsoft, Google, or IntelliJ stealing other people's code with an IP laundering machine.
Add comment