3 ms·
Today I already have trouble doing git archeology to understand why a piece of code was done in a certain way. Now imagine debugging more and more code that wa
by gtirloni 3y ago
Today I already have trouble doing git archeology to understand why a piece of code was done in a certain way.
Now imagine debugging more and more code that was created by LLMs.
Debugging will become more of a voodoo witchcraft kind of thing. I pity anyone who decides to make a living out of that. Probably will earn a lot and die young.
- BigJono 3y agoPeople are very quickly going to find out that "never rewrite" can't be taken as absolute advice. If you have 5Mloc of third party dependencies and AI generated code that you can't extend and can barely keep functional in production, and a competitor pops up with 50Kloc nearing feature parity and adding more functionality by the day, you'll very quickly need to adapt or die. Software companies already die to this risk aversion all the time. Tons more die to rewrites because they get conned into hiring an army of idiot React consultants to do it, who are likely in these HN threads at the forefront of AI code gen themselves, and just end up making 2.0 more complex and worse. All GPT does is widen the gap between good and bad engineering. If it widens it enough that non-technical people can finally tell the difference, then it's going to kill half the market, just not the half everyone thinks it will.
- manicennui 3y agoI imagine that it will be about as bad as debugging code written by the sort of people who find "AI" helpful.
- 62951413 3y agoAren't we going to get copious amounts of documentation for free? To me chatGPT logs look like JIRA descriptions, commit messages, and basic design specs in disguise. And they are not optional anymore. On a related note, I'd expect the "DSL+generated code" approach to grow in popularity for common domains. Think gRPC or JOOQ. A terse IDL snippet and you get a lot of functionality that you don't even need to test or even put under version control.