7 ms·
The way you tackle the unreliability of coding assistants is to know what you're doing so that you don't need a coding assistant, so that you're only using it t
by kazinator 3y ago
The way you tackle the unreliability of coding assistants is to know what you're doing so that you don't need a coding assistant, so that you're only using it to save time and effort.
Roughly speaking, if you stick to using AI for writing code you could have written yourself, you're okay.
- NateEag 3y agoThough regular use of the assistant will degrade your ability to program without it.
- majormajor 3y agoI think in the same way that autocomplete does - I may not have typed out `.length` or whatever it happens to be in the language I used last year enough to remember a year later if it's len or length or size and if it's a property or a method after a year goes by without me touching that language... but then it's a simple search away to refresh my memory, and overall the autocomplete saves a LOT of time that makes up for this. Yeah, if you never knew what the code that got generated did in the first place, that's not gonna apply, but if you're using it as basically just a code expander for things you could do the pseudocode for in your sleep, you're probably gonna be ok.
- digging 3y agoI've only used a very small amount of AI assistance (mostly Anthropic's Claude), and always to learn, not to do. That is, I will ask it what's happening, why are things breaking, etc. It doesn't have to have the right answers, it just needs to unblock me. I hear it is also quite useful for doing things which you know extremely well but are tedious to do. Anything in between is certainly a danger zone.