I use LLMs in much the same way, but lately I have been wondering how good I will continue to be at thinking, verifying, or communicating in code, if I stop writing as much of it. I suppose time will tell...

I’ve been using Copilot since the beta and it does cause dependency, which can be dangerous because you have to be constantly vigilant of errors it might be introducing. It’s a good idea to turn it off sometimes and try to solve problems manually.

I highly recommend any programmer seriously using ChatGPT to go through this entire guide:

https://learnprompting.org/docs/intermediate/chain_of_though...

Providing context, getting GPT to explain the steps before doing them, creating feedback loops, etc is super helpful when doing this stuff.

It's still useful to just go "write this function in x language" but anything beyond a Stack-Overflow type copy/paste script it helps to do a lot more hand-holding. The ROI is worth it.

do you have some examples you could share? where y9ou were successful and it was worth the ROI?

It is very nice not having to write code from scratch by hand, but I have noticed that by not doing so, my innate familiarity of the code absolutely craters. Making manual changes to the code suddenly becomes a much more tedious and less confident process, and in many cases I am not sure that it's a net benefit (at least in a situation like mine, where there is no oversight and things go straight to prod.)

Same for me. I wonder if this is going to follow a similar pattern that we went through with C vs. Assembly. There was a period where everyone coded in assembly and then C came along but people still hand-coded certain parts of applications in assembly to make them more efficient. Then eventually when C compiler optimizations got good enough and computers got faster, few people bother with assembly.

If it is not doing so already, ask it to comment the code it writes.

Worst case scenario, throw the code back into GPT and ask it to make the changes you want to make.

My take-away from your take-away is that LLMs might well take away the one part of my job that I genuinely enjoy doing. Writing code, by hand, myself. Reasoning from first principles.

This is also one of the reasons I transitioned to developing AI models myself. At least I'll be safe until they start rewriting themselves from the ground up. And at that point we'll all have bigger worries to deal with anyway.

Yeah, you should be safe for at least another few weeks :)

I used to get the same results, but I've lately been using better prompts and the results are substantially better.

The key is to a) give it context, and b) give it constraints, and to do it from the very first prompt itself.

Most chatGPT failures that I've seen add context and constraints after the initial prompt, which someone sends GPT into a loop.

If you have data in a particular format, mention that in the first prompt. Also helpful to tell it to optimize the code for scalability and to use any external libraries if necessary.

These tools are essentially software programs, and like any piece of software, you need to know the right commands to get maximum utility out of them.