https://github.com/anthropics/claudes-c-compiler/issues/1
-
@GossiTheDog Sloppy, no sloppying!
-
@GossiTheDog "Claude will work autonomously to solve whatever problem I give it. So it’s important that the task verifier is nearly perfect, otherwise Claude will solve the wrong problem."
Let me get this straight. We task the AI to program for us, but in order for that to work we have to spend time writing near perfect test-cases in order for it to work.
What's our experience with defining "exact" and "near perfect" test-cases? Exactly. We are near perfect at being rubbish at writing near perfect test cases.
I look forward to this future.
-
@GossiTheDog
Hahaha... 
-
@GossiTheDog Terminator's DoomsDay won't be caused by SkyNet becoming sentient, but random AI "agent" making "a mistake", whoopsie !
-
@GossiTheDog Minus several million out of ten.
The whole point of a Hello World is to >>sort out all the extraneous bollocks about which you so far know nothing for you<< and give you something that Just Works, that you can then edit to do something more useful.
It had one job. It failed.
-
@GossiTheDog so they tasked Claude to create a C compiler using Rust. Surely to be valid they first must have used Claude to build a Rust clone?
-
@GossiTheDog I almost feel bad for them because of how much of a shitshow this thread is
-
R ActivityRelay shared this topic
