GLM 5.2 is the first open weights model I'd call a genuine competitor to Opus and GPT for agentic work - at ~15-20% of the price. Part one of why AI inference margins are about to collapse.
A friend of mine took the attitude of using it at work where it may be useful at speeding things up, but for personal projects “FUCK NO, I’m not putting a single line of AI code or assets or anything into my stuff”.
that’s a good call. vibe coding on drugs is a pretty chill futuristic time, but there’s no way I’m putting any of the results out there like I’m proud of it. The only things I accomplished were getting high and creating homework for myself to look at later.
I wouldn’t rec AI tools to a novice, that’s for sure. You can’t really build a good intuition for programming if you don’t go in there and write the programs yourself. the higher-end models can do a lot, but if you don’t understand what the code is doing on your own the result is always going to be slop.
I treat the AI as a pair programming partner that’s reeally good at pattern matching, but has zero comprehension skills. I find this makes the tool way more useful since I’m fighting slop and nonsense way less. the hard part is training yourself to break the programing problem down into small auto-complete steps that a medium strength AI can run with. basically my approach is writing a program or patch in my head and prompting the AI to generate the components.
for hobby coding, this is way more work than just writing the code myself. it only pays off when dealing with large projects. its like an industrial power tool, useful in-context, but you don’t need a hydraulic power-hammer to do a little carpentry.
Honestly, If the VSCode auto complete with AI was just a little less zealous, I would love it. Other than that, I don’t think AI is worth it for coding. Maybe small things that you would have to pour over documentation to make one thing otherwise, but especially a large project, its more hassle than its worth.
my concern for myself on the occasion that i do this is that i get worse at coding every time i do it
A friend of mine took the attitude of using it at work where it may be useful at speeding things up, but for personal projects “FUCK NO, I’m not putting a single line of AI code or assets or anything into my stuff”.
that’s a good call. vibe coding on drugs is a pretty chill futuristic time, but there’s no way I’m putting any of the results out there like I’m proud of it. The only things I accomplished were getting high and creating homework for myself to look at later.
Makes sense lol you sound pretty down to earth about it and I think among tech oriented people it’s starting to fall into these sorts of places.
I wouldn’t rec AI tools to a novice, that’s for sure. You can’t really build a good intuition for programming if you don’t go in there and write the programs yourself. the higher-end models can do a lot, but if you don’t understand what the code is doing on your own the result is always going to be slop.
I treat the AI as a pair programming partner that’s reeally good at pattern matching, but has zero comprehension skills. I find this makes the tool way more useful since I’m fighting slop and nonsense way less. the hard part is training yourself to break the programing problem down into small auto-complete steps that a medium strength AI can run with. basically my approach is writing a program or patch in my head and prompting the AI to generate the components.
for hobby coding, this is way more work than just writing the code myself. it only pays off when dealing with large projects. its like an industrial power tool, useful in-context, but you don’t need a hydraulic power-hammer to do a little carpentry.
Honestly, If the VSCode auto complete with AI was just a little less zealous, I would love it. Other than that, I don’t think AI is worth it for coding. Maybe small things that you would have to pour over documentation to make one thing otherwise, but especially a large project, its more hassle than its worth.