Will ChatGPT replace programmers?

It was mentioned on our company Slack yesterday, and since I had the day off but was stranded at home while our new porch’s roof was being installed, I spent a few of hours playing with it.

I made repeated attempts to get it to generate code I could use directly, but never once did it manage to do so, at least not without more work than if I had written the code from scratch. It produced examples that, at best, gave a good notion of what an assignment required without giving away the actual solution, similar to those I once offered in university classes.

It could be useful for creating regular expressions. It produced results after I requested numerous common examples and a few unique problems. I’m not sure if they were any better than using a standard regex generation tool, though, or if it wasn’t just as much labour.

Also, I’m not sure whether I’d believe that to be accurate.

I requested the last digit of pi from it. It asserted with confidence that it was 9, but with a few fair qualifiers, and then provided me with an example of pi that had around a dozen numbers to back up this assertion. The score was 6.

Then I requested many interviews with eminent historical individuals and various fictitious pieces on specific subjects. The results were all theoretically competently written; the grammar was correct and they were simple to read; however, they were all dull and superficial, as if they had been produced by a clickbait content creator or a bored high school student who didn’t care about the subjects being covered or the actions of the historical figure.

I enquired about how to start an antique motorcycle. The response was partly true, but it completely omitted the crucial information that would have helped prevent injuries from improper use of the equipment.

A coworker made a comparable inquiry for guidelines on how to land a glider. He claimed he believed it was attempting to murder him.

Natural attempts at communication were unnatural, uninteresting, and devoid of enthusiasm. It resembled a poor dinner date with a person who didn’t like me and was a little bit severe, humourless, and judgmental.

Speaking of dinner, it generated recipes remarkably well. Requests for vegan stew that included chicken were properly rejected as contradictory, but (let’s say) a request for vegan stew that included lentils and potatoes resulted in what appeared to be a recipe for a tasty, though uninspired, dish. It might be a little monotonous, I think.

Even so, I’d be happy to utilise it for that.

Overall, I got the idea that it was what we could reasonably anticipate from a modern-day ELIZA, but like ELIZA, it’s still just a toy, even if it could be the basis for something like Google Assistant or Amazon Alexa.

replacing programmers, though?

No, not even close. I’ll be concerned when it can conduct user and stakeholder interviews, design a solution in accordance with company direction and corporate vision, while taking into account budget, timeframe, preferred and available infrastructure, industry regulations, and current technology trends, as well as meeting functional and non-functional requirements.

It doesn’t even seem like it could be a competent programmer’s assistant at this point—a capable mechanical coworker who can rapidly do the tedious, repetitive sections so I can concentrate on the interesting parts.

Before we receive that, I believe we’ll have to wait for GPT4 or GPT5, or perhaps later.