I mentioned it in a previous comment. It's common for senior engineers not to write that much code. They spend time their time on meetings, planning, creating architectures, presenting solutions, discussing solutions, triaging, keeping up-to-date with tech, clarifying business cost, working on waste avoidance, reviewing code, streamlining processes, vetting new tech/solutions and, in general, understanding everything that is going on.
ChatGPT+ will definitely have some affect on junior devs but us more experience folk should be fine... for now..
One thing about GPT is that it only knows what we know at the moment. That indicates to me that it won’t be great for learning new technologies until humans generate content it can regurgitate. That alone might give juniors an edge against it (assuming they are gradually replaced by a robot pooping out dumb logic) - they might be able to specialize in learning what models don’t yet know, or what they can’t be good at.
Just guessing here. I’d love to hear a rebuttal to get a sense of where people think things are going.
Though I don’t think GPT is “there” yet, I can see it getting there by 2030. I think it’s seriously worth considering: how will people learn to program in 10 years, how will they remain relevant through periods of their career where an AI can generate better solutions than they can, and how will more experienced engineers adapt to those changes?
I like to think of Industrial Revolution for counterexamples. It wiped out tens or hundreds of professions. No more blacksmiths in every village.
Two things happened: one, blacksmithing effectively was taken over by a different skill, that of configuring, servicing and operating industrial machines. There were no junior or senior blacksmiths anymore, seniors probably migrated their knowledge, and the whole field was taken over by non-blacksmiths.
The other thing is that a small proportion remained. Initially perhaps for niche goods that were too difficult or uneconomical to automate. Now blacksmiths do exist, a bit like craft coffee blenders and horse trainers. But the industry was nonetheless wiped out.
Will we see that crafting software will become a job for people trained in AI whispering? Perhaps they will never, or hardly ever, write any code, but will train extensively in software design. Much as a metallurgy engineer might know all crystalline forms of steel but have never swung a hammer.
Day 1 these jobs would likely get filled by senior devs, but there may be no new devs, junior or senior, except for niche applications.
I guess they won't, in the same way we are much worse at remembering information in the era of search engines. They will be profient in telling the AI the exact words it needs to produce the correct code output. No one will be good in programming trivial code snippets not because every snippet is on SO but because every code part can be generated with the correct request.
As many of us cannot read machine code and hex anymore many of the juniors won't be able to parse the code output - but it may not matter anymore.
They will probably manage. People manage to become programmers without having to learn how basic electronics, cpus, operating systems, etc work. You just skip over those solved problems.
> ChatGPT+ will definitely have some affect on junior devs but us more experience folk should be fine... for now..
It affects both. If a single team was to be split with 5 juniors and 5 seniors, ChatGPT significantly reduces that headcount from 5 juniors to 0, and 5 seniors to 2 or 3.
With many companies cutting costs and with the cheap money getting dried up, no-one is safe. HN may not like it but, the same thing that has happened to digital artists with Stable Diffusion which was welcomed on this site, now has happened to programmers and I see lots of worrying and frowns everywhere.
It appears that StackOverflow (which lots of juniors and senior developers use) has just become irrelevant.
> ChatGPT significantly reduces that headcount from 5 juniors to 0, and 5 seniors to 2 or 3.
Citation needed. I haven't heard of any massive disruption in the commission art market since stable diffusion went public, and I don't think something less-impactful(a different way to search old stack overflow posts) is going to cause a massive disruption either.
Stack overflow still beats chatgpt in one area that it can never compete. Coming up with new solutions to new questions. If all we needed answered was the same questions, chatgpt would be sufficient since it's essentially a compressed version of our current knowledge. We don't really have a way to update it with "new knowledge" other than "train it again".
Art is a strange market, where people pay for more than the physical good being delivered. In some cases the objective value of the good being delivered is zero, or even negative after considering cost of "carry" (transport, storage, insurance) yet people pay for it.
I once spoke to an illustrator who said her work was definitely shrinking due to low cost human competition, so much so that she was leaving her profession. I can imagine this did in fact get worse.
I think the opposite, LLMs will be used to build the optimal high level scaffolding and implementation, but low level devs will be needed to check and verify the code. As we ve seen so far, AI automates the brainy part, but not the long tail or parts that need physical access (eg safety drivers, warehouse workers)
The point is exactly that most of those meetings are happening everywhere for the same reason and thus GPT25 might already know all the answers that you need.
Also given enough general framework skills, I'm pretty sure the AI will be able to build stuff like a good junior dev.
The algorithm doesn't have any "general framework skills" though because it's an algorithm, not a person.
It can generate something that looks like what a person would have wrote based on its compressed probabilities, but that's very different from being an "an artofficial intelligence". At best it's a Chinese Room.
And proper APIs and Frameworks are very well documented inputs and outputs, whats your point?
Im pretty sure we could train an AI to suggest the correct API for the intended purposes. Given the fact that we have support of datatypes and documentation what is left to do is to map what you have to what you want.
Also we can reasonably generate classes and controllers already through cli, so why not combine all the nice things we have to something usefull.
With types and comments as a prompt in Go, GPT already produces fairly viable API endpoints for basic CRUD operations.
With some framework to compose various prompted endpoints using the same types and conventions, a junior engineer would be totally unnecessary. They couldn’t code as well, and they couldn’t architect as well as a more experienced engineer doing the prompting.
You could even prompt from a spec and begin iterating upon the outputs, revising each manually or with GPT’s suggestions. Say “take this endpoint and refactor it based on the conventions of this endpoint, and make sure (some logic) accounts for (some potential problem)” and boom, it will probably work okay.
But aren't the devs of the future on GPT right now, half-assing learning to code while getting a fast track education in correcting bugs and architecting?
If this technology evolves to be able to reliably generate working code to a prompt, the entire field of software dev will shift dramatically. Some junior and some senior devs would prove better at meeting deadlines by AI whispering, and others would have to find new career paths.
ChatGPT+ will definitely have some affect on junior devs but us more experience folk should be fine... for now..