Will ChatGPT Replace Programmers?

It was brought up on our work Slack yesterday, and having the day off but stuck at home whilst the roof was being put on our new porch, I spent a couple of hours playing with it. I made numerous efforts to get it to produce code I could personally use, and not once did it produce anything that was remotely usable, at least not with more work than creating it from scratch myself. At best, it produced examples of the sort I once provided in university classes to give a good idea of what an assignment required without giving away the real solution. Where it might have use is generating regular expressions. I asked for several standard examples plus some custom problems, and it generated results. But I'm not sure they were better, or that it wasn't more work, than using a conventional regex generation tool. I'm also not sure I'd trust it to be correct. I asked it to give me the last digit of pi. It confidently, though with some reasonable caveats, claimed it was 9, then gave me an example of pi with a dozen or so digits to demonstrate this fact. They ended in a 6. I asked it for several fictional stories on given topics, then several interviews with famous historical figures. The results were all notionally competently written ? the grammar was correct and they were easy to read ? but dull and superficial, like those written by a bored high school student or a clickbait content writer who knew what the topics were about or what the historical figure did, but didn't care. I asked it for instructions for starting a vintage motorcycle. The response was accurate, mostly, except for entirely leaving out the important bits that would avoid the risks of injury from misusing the kick-starter. A colleague did a similar request for instructions on how to land a glider. He said he thought it was trying to kill him. Attempts at natural conversation seemed stilted, dull, and joyless. It felt like a bad dinner date with a slightly stern, humourless and judgemental person who didn't like me. Speaking of dinner, it did surprisingly well at generating recipes. Requests for vegan stew with chicken were appropriately rejected as contradictory, but (say) a request for vegan stew with lentils and potatoes produced what looked like a recipe for a pleasant, though perhaps somewhat uninspired dish. I suspect it would be a bit bland. Though I'd happily use it for that. My overall impression was that it's what we should expect from a modern-era ELIZA, but like ELIZA it's still basically a toy, though it might make a great core for something like Google Assistant or Amazon Alexa. But replacing programmers? No, not by a long shot. When it can interview users and stakeholders and design a solution in light of company direction and corporate vision, whilst considering budget, timescale, available and preferred infrastructure, industry regulations and current technology trends and meet functional and non-functional requirements, then I'll be concerned. For now, it doesn't even look like it could be a useful programmer's assistant; a capable mechanical co-worker that can quickly hammer out the repetitive dull parts so I can focus on the good bits. I think we'll have to wait for GPT4, or GPT5, or later, before we get that. Read Also : How does a compiler engineer compare to a regular software engineer? Technically, day to day job, skill set etc.