406 karma · joined March 9, 2016
Maybe I write a bot that answers fivverr requests at the lowest price possible. We can all race to the bottom.
Isn't this a problem with the agent loop / structure, rather than the llm, in that case?
The ide doesn't affect the models results, just what is done with those results?
What does chef do? Garden? Pig? Burp?
Nonsense.
Several projects later, I can work with the AI now to produce several thousand lines of code, I understand it, it is what I asked for, has the architecture I wanted, is not bloated and is not buggy.
Spend time planning, figuring out the architecture you want, refining, breaking things down into smaller implementable features with implementation plans for each feature. Review these things aggressively. Make sure they align with your expectations.
Then get it to start writing code. Writing tests for everything. Include detailed runtime logging. Read what it produces, understand it and redo it if it wasn't what you expected. Your experience will be vastly superior to that first attempt.
You can save time still, but perhaps not as much as you think, because you need to check the ai's work thoroughly.
It didn't produce anything close to a competent go playing program, but it did get to the point it would play legal moves, which considering I started it with a purely randomly generated population of programs, it had zero built in knowledge of the rules of go, and had it playing against a well known open source go playing program to assess fitness of a given pop, I was quite pleased with it.
This was around 10+ years ago, just as Google made alphago.
We have a guy who joined recently who constantly tries to be as clever as possible when writing his code. He ignores established patterns and conventions because he knows better. He does some cool stuff. Really cool stuff.
He's so far bounced thorugh 3 different teams, I don't think he'll be with us long.