colwont

Yup. This is the only post that hasn't been marked dead :D

This is something I've asked for the best part of a decade and one of the causes is the disconnect between test and development.

So to solve this I've been developing `ccc`, a static analyser & runtime instrumentation tool to determine: [for any given change; what is the minimum required test to cover that change].

Preliminary benchmarks and testing for a multi-component SaaS with a Python Test suite (integration, perf etc) have shown positive results. Test suites requiring full-workflow runs (100s of tests) reduced to single or a handful of smaller faster tests.

Looking for feedback and to hear if anyone working on large code-bases experiences similar problems in their projects.

Cheers!

1. A specialised system is not super-intelligence; it's a specialised system.

2. Speed is a terrible measure of intelligence.

3. Manufacturing processes' are just as intelligent as LLMs.

4. Output from a specialised system that is non-deterministic and unable to generalise is not quality output.

I think this dude has made a sum amount of money as an anti-AI grifter and is now flipping their stance to cover a wider audience.

I was wondering this too, I worked on a project for weeks, posted on HN and got shadowbanned lmao

undefined behaviour

Was also going to say this. If anything, the borrow checker in Rust is more likely to save you from LLM issues, because it won't bloody compile.

nailed it

yup, show receipts

I think I'm shadow banned... :(