the only software developer thing chatgpt does well exceptionally well is 101 level answers to general questions/requests and read to me paraphrased stackoverflow results with nearly google levels of reliability.

Where chatgpt really 100x's a person's output is when you're trying to generate shitloads of spam text, such as automated posting of unique comments that use the post/thread/blog/videos's context and existing comments as context to appear relevant while still pushing a narrative or shilling a product, or building a proxy such that every page someone visits on your website, you automatically reword (plagiarize) another specific website's article then add your ads.

I don't work in actual software development, though I do a little of it amongst other work.

When I need to slop out a one-time snippet or short script to do something, which I have to do like 10 times a day, it takes me like 3-20 minutes. ChatGPT 4 does it near-perfectly, takes one minute, and usually teaches me something on the way.

Plus when I need to work out how the fuck GDB works to debug shit, it's an absolute lifesaver. The manual is very long and remembering all the memory examination commands is hard.

If you're ever working on code over ~100 lines a long, then I basically agree as it takes massive debugging and is poorly factored to the point of being worthless. But for arcane, well-documented commands (ie obscure programming languages and linux tools), and short blasts of code, it's genuinely incredibly useful on a daily basis.

Idk you probably sound like people did when search engines first started getting popular. If you can’t learn how to get good output from an LLM you might get left behind. I never use LLMs for large chunks of code just snippets and it’s great for that. It’s just like StackOverflow. Don’t blindly copy shit without understanding what’s actually going on. You have fun writing boilerplate code I’m never going back to hand writing that shit.