Ai: “I see the issue, let’s fix it. It’s been fixed “
Me: no it hasn’t … here is where it’s failing.
AI: ah, you’re right, I see the issue, let’s fix it. Now it’s fixed.
Me: no, still not fixed, here is how you’re doing it wrong.
Ai: my apologies. I see the issue. Let’s fix it. There it’s fixed.
No, it really isn’t. 🤦♂️
And then while it might get fixed, it’ll then quietly remove another feature without telling you and then you’re all confused until you debug and see it missing
🫂
That’s when it’s good to ask the AI to write a CHECKLIST.md file with the items it needs to complete to be a success and refer to it before attempting the next fix.
Hmmm. Does this work?
This is from a thread I posted about it on Twitter. It’s how I’ve been coding for about a month now … it’s not perfect BUT it’s much better than before
For reference here’s an example of a checklist … but I also usually add in things like code strategy or session goals or best practices like telling it to use ESModule for all Typescript code and not convert to CommonJS even though it might have memories or reference code that it will want to write in that format.
Reminds me of apprentices 🤣
You are a helpful assistant
Yes I am
"If you need something else, tell me"
I had this exact experience except then it went "This is going to take a while. Please let me work on it & check back with me later."
It was a few hours before I realized it doesn't do background processing or anything like that & so it's basically just lying to me & telling me it's working on something it isn't. And, people say AI can't replace humans!
Did you know that the G in ChstGPT stands for Gaslighting?
Clearly!! Thanks for pointing it out.
can you expect a more patient assistant
AI is the maximum bullshitter.
For a similar behavior in the physical space, people has been rightly punched in the face
Exactly this. Although as a noob I think it has made me better at reading code & debugging.
Meanwhile people some people still say it’s AGI
🤦🏻♂️⚡️💜
I (no coding experience) was hoping to build an app with replit’s ai, but this was sadly my experience as well
Too familiar. On top of that AI tends to blindly trust too much.
Me: it ain't working because of this, let's do it that (still wrong) way instead
AI: you're absolutely correct! let's do it. here you go
> shit still broken
Me: oh my bad, that was a stupid idea, that would never work, let's do it this other, still wrong, way
AI: I agree with you! that was not the right way. let's fix it
> does what I say even if it's blatantly wrong
> shit still broken
