I much prefer a world where people can look under the covers and understand what's going on, rather than leaving it to AI. That's what a healthy relationship with AI would look like.

Computer programmers will be replaced by software? Where did I hear that before...

I'll be keeping my text editor, thanks.

Computer Science is not (only) programming. And judging from ML generated output so far, I don't think we should just trust the system to give us correct results. I would say that journalists (who hopefully do check sources!), developers (and that includes much more than just writing code, ffs) and scientists so far are pretty safe.

Learning to think has never been a bad thing. I don't write(a lot of) code any more, but it certainly gives you an insight on how to break problems into smaller pieces while not losing sight of the bigger picture. Same with hand tool (woodworking is most accessible) skills and needlework. Learn a bit of both. It is not only fun, but does challenge your brain in a wonderful way. My kids can sew and saw in a straight line.

He said CODERS would be replaced by software.

That I can believe (eventually, some future generation of AI well beyond the current one) because that's basically rote translation of a concept into code. That's easy, a lot more people can do that than the truly difficult part of programming. That is, gathering requirements and creating a spec for what the program is supposed to do. That's the really hard part, and AI won't help you there - but he wasn't saying it would.

But assuming some humans have done that and produced a detailed spec saying exactly what the program should do and how it should react to all the various inputs and circumstances then yes I could easily see some type of automation take over and translate that into whatever computer language you prefer.

It won't do what you want, because you'll have got the spec wrong, but the one improvement of this method of programming over the current one is that instead of making code changes trying to fix the shortcomings/errors in the spec (i.e. do what the end user WANTS rather than what he told you or what he thought he wanted until he saw it in action and realized that wasn't what he wanted after all) is that after fixing the spec you can have the automation generate new code from scratch. Basically the spec (which might be part English, part flowcharts, part mathematical formulas, part government laws and regulations, part APIs from third parties, and who knows what else all combined into a toxic witches brew) becomes the source code.

I've long thought one of the reasons there are so many computer languages is that academicians think programming is writing code, so they think a language that makes that code more "elegant" will be better. They don't understand what real world programming is.

Ah, yes.

You are correct.

Just soooo many people confuse the two. I assumed that ... id-ten-T was one of them (still 90% certain)

The academics are not programmers. The academics are more theoretical and do get the algorithm part of computing science. They are not at fault for the bajillion different languages. And, to me, all languages in a certain group are same-ish. C, FORTRAN, R, Basic, Pascal all are procedureal. Then there's object oriented, and finally the weird Lisp stuff. Most things that divide languages in a group is syntax and some details (like row major / column major, pass by value / pass by reference, which is important, of course, but not enough to make them oooh so different).

but the one improvement of this method of programming over the current one is that instead of making code changes trying to fix the shortcomings/errors in the spec (end user spec) is that after fixing the spec you can have the automation generate new code

If that were true the end user wouldn't need you as a middleman. That's what CEO's dream of, and promise their activist investors in exchange for larger bonuses. From that POV, AI is the new outsourcing, a good excuse for getting rid of or demoting anyone who isn't in business and sales. How's that going, Boeing?

Basically the spec (which might be part English, part flowcharts, part mathematical formulas, part government laws and regulations, part APIs from third parties, and who knows what else all combined into a toxic witches brew) becomes the source code.

AI is nowhere near that. Where it is now is another iterative improvement. Hard wiring replaced by punch cards, replaced by microcode, replaced by complied languages, added to by interpreted languages, enhanced by smart editors. AI is another programmers tool. It's limitations are just as apparent as it benefits. For those limitations to change by another few orders of magnitude is going to have to wait until the next hardware revolution, because of resource barriers.

... eventually, some future generation of AI well beyond the current one ... Maybe - meteor strikes, etc. permitting. But fuzzifying what is with what might be is not engineering, it's entertainment or dreaming (fine), or a psuedo-religion (ugh), or in the case of Jenson Huang and some others, good sales talk (expected).

OK, so you change the spec to fix your bug and regenerate the code from scratch. And now you you can throw away your carefully collected and filtered bug list because those bugs are gone and a bunch of new ones have popped up!

Fix the next bug in the spec, regenerate the code again and.. Oops, same thing happened! Bin the list, start QA again, repeat. Possibly forever.

Software always starts wrong, and (sometimes) eventually becomes good through a process of incremental improvement. I don't think regenerating code in a non-deterministic way from a spec document can really be a good substitute for that process.

The other thing I notice in your example is that the spec now has to be massively more detailed and precise than a spec has ever been before - probably so dense and complex that a non-technical person cannot understand it. Now it's.. the code!

Your "detailed spec saying exactly what the program should do and how it should react to all the various inputs and circumstances" *is* your computer program.

It sounds like you are just inventing a new programming language. It might be better than existing ones, but it is still a programming language.

I remember back to at least 1981 or 1982, when software called "The Last One" was released, claiming at the time that you'd never need a programmer again.

It goes back to well before that. That was the aim of Cobol and its predecessors in the 1950s.

I've been hearing the same story for 40 years or more. I can still vividly remember my dad telling me not to become a programmer in the early '80's when he showed me an article on so-called Expert Systems. People would tell them what they wanted and it would produce a complete working program as if by magic. We all know how that fanned out.

I believe there's a deep desire among companies to replace these expensive and obnoxious software developers, as we can see with all the hypes (Expert Systems, Lo-Code, AI). It will eventually become reality but only when computers have matched human beings in intelligence. And I believe we'll have different problems on our mind when that happens!

Coding was the first thing outsourced offshore decades ago. It hasn't been a good career choice for half the world for ages.

Coding is the process of telling the machine what to do. First however you have to understand what and how you want to tell it, and that's the tricky bit.

There is a new IT discipline called Prompt Engineering, which people are making a lot of money out of. Prompt Engineers develop the natural language questions / instructions you enter at the AI/ML prompt. They can even develop template prompt texts which you can vary certain parts of (you might call these variables user input). All of a sudden, Prompt Engineers look rather like software developers.

Isn't it funny how they never seem to think that business management can be replaced by LLMs?

Actually, some layers of management are good candidates for replacement by LLMs. I suspect others already have been replaced by software: e.g. I know software decides routes in logistics, that must have been at least partly a junior to mid management decision before that.

There have been computers doing financial trading for a while too.

... Let them grow up to be gamers, to provide a stable income stream for nVidia.

>"The technology divide has been completely closed."<

Dunning Kruger in full effect?

If you don't know how to interpret the code the LLM comes up with how will you spot it when it provides something that is likely according to the training data but still wrong?

You could ask another LLM what's wrong with it. With the proper tools (and prompt engineering!) you could even let them talk between themselves, discuss the code, fight about indentation and language preferences, disagree about variable names, and just stop working together because their partner is an a**hole. Just like real meatbag programmers!

Then you can add a third LLM to act as the group's leader and a fourth as project manager and see what happens. It'll be LLMs all the way down!

At the moment machine learning makes everyone a questionable programmer, depending on how well they can find flaws and hallucinations in the generated code. There may also be copyright issues.

So will the contract rates for skilled programmers who get desperate phone calls to come in and sort out the total fucking mess made by anybody stupid enough to actually buy into the idea that he's selling.

I have literally been hired as a programmer for my ability to drive a counterbalance forklift truck so the company didn't have to hire someone for one hour per month.

IT (Programming, Infrastructure, Security ... everything) is the art of giving the end user/customer what they need, despite being told what they want.

I'm sure he does. He's an electrical engineer, not a programmer. Obviously, he thinks that, as long as you can find some code on the Web, you don't need a programmer.

Unfortunately, there is no pseudo-AI that is going to code Salesforce, or SAS, or an ERP. I don't care that there are now "prompt engineers", you still have to know where to place the prompt in the code.

Programmers are not going away any time soon.

Both?

'Coding education' is strictly training , and is the most miniscule element of computer science. Computer scientists set the parameters for the design of computers and solutions to computing problems. Coders implement already defined algorithms using specific languages. Both (together with other skills such as algorithm design) are necessary contributors to systems and software development, but one of the fundamental reasons why current mainstream computing technologies are such crap is that almost everyone thinks computer science, software development and coding are all the same thing -- defined by coding.

Quite apart from which, if neither is to be taught, who will be competent to check that the AI hasn't generated garbage, let alone fix its cock-ups?

...company infamous for price gouging when their products reach a dominant/essential market position, encourage society to de-prioritise skills that would decreasee reliance on their product.

Think I'll pass on that gem of advice Jensen.

If you keep training computer scientists then one of them may come up with a better way of doing "AI" that doesn't require our expensive hardware