Post B7WUVzNWH5vMWgn9Jg by anthropy@mastodon.derg.nz
 (DIR) More posts by anthropy@mastodon.derg.nz
 (DIR) Post #B7WUVzNWH5vMWgn9Jg by anthropy@mastodon.derg.nz
       0 likes, 0 repeats
       
       saying "LLMs often generate bad code" doesn't really land with the people you're trying to reach because the same could be said about humans. it's a moot point.emphasise its use as a (optional) tool instead; you can't replace a carpenter with a nailgun either, you'd need to understand what you're doing.No middlemanager is going to be able to build the next Kubernetes using LLMs, because they don't even know what a container is to begin with. Use that.
       
 (DIR) Post #B7WUVzYrauzx5s6Czw by enigmatico@mk.absturztau.be
       0 likes, 0 repeats
       
       @anthropy@mastodon.derg.nz It's the small details that matter. Humans can produce bad code, but humans learn from their mistakes and correct them. Machines can't, you need to give them actual code for the machine to be able to spit something that somehow works, and even then it misses a lot of points. You can correct the code and make it work, but then what's the point of using AI if you need human intervention anyways?I remember someone comparing it to FPSs in games. A company can make a game that runs fine, but at 30 FPSs, using AI. Or they can make a game that runs at 120 FPSs, optimized and engineered by a human that know what they're doing.And then there is the problem of maintainability. A machine can spit out a lot of code, but then after a few years if you need to make adjustments, who is going to examine the code and fix it for you? Another LLM? And if the LLM fails then what?