No, don't use a AI for anything serious, except if you have absolutely no clue what you're doing and you just want to know what a piece of code is doing. And even then it's better to ask a human than a large language model. As the person in the video noted, if you use a LLM to generate code, you should know what the code is doing. Then, you need to take out the bad, buggy code and replace it with something that works. The end result: you've spent more time reading and fixing an LLM's code and you haven't thought of a meaningful solution yourself, meaning you'll feel less satisfied with the solution. One exception to the rule is if you hate the creativity that comes with programming and love the frustration that comes with debugging. Then absolutely use LLMs when programming.
I have found that AI seems to emit relatively few bugs (certainly fewer bugs than any random human developer), but it's great for doing tasks you've done many times before and won't learn from doing. Debugging can be extra frustrating indeed - but learning how to work efficiently with the AI output is a skill itself I do think there's something to the idea of stagnating your growth by leaning too heavily on it but I also think it can have the opposite effect, empowering developers to move MUCH more quickly and build more interesting projects to actually learn about interesting things
@@johnk6757 But if you've already done something, then surely you've made a component out of it? I can't think of anything that can be automated with LLMs but not with traditional code.
No, don't use a AI for anything serious, except if you have absolutely no clue what you're doing and you just want to know what a piece of code is doing. And even then it's better to ask a human than a large language model.
As the person in the video noted, if you use a LLM to generate code, you should know what the code is doing. Then, you need to take out the bad, buggy code and replace it with something that works. The end result: you've spent more time reading and fixing an LLM's code and you haven't thought of a meaningful solution yourself, meaning you'll feel less satisfied with the solution. One exception to the rule is if you hate the creativity that comes with programming and love the frustration that comes with debugging. Then absolutely use LLMs when programming.
I have found that AI seems to emit relatively few bugs (certainly fewer bugs than any random human developer), but it's great for doing tasks you've done many times before and won't learn from doing. Debugging can be extra frustrating indeed - but learning how to work efficiently with the AI output is a skill itself
I do think there's something to the idea of stagnating your growth by leaning too heavily on it but I also think it can have the opposite effect, empowering developers to move MUCH more quickly and build more interesting projects to actually learn about interesting things
agreed
@@johnk6757 But if you've already done something, then surely you've made a component out of it? I can't think of anything that can be automated with LLMs but not with traditional code.
i dont think i can run an llm locally on my 2019 macbook with a whopping 4gb vram
Yeah, you're cooked, even with a RTX3080 I am struggling to run one!
Like them or not, you will be left behind if you don't learn to use them efficiently.
Ok