Ask HN: Is HN in denial about the impact of AI?

[ThrowAIAway]

I've been lurking here for over a decade and love this community. I'm 50 years old and not a programmer but work in IT.

I saw a few discussion about the impact of AI on developers on HN. It seems, while not a consensus, a lot of folks here think this LLM thing is overblown and will somehow go away or the AI progress will not be as bad as everyone is trying to make it to be. While others are pointing out that the job of a developer is not all about coding, but gathering requirements, understanding customer needs and then coding. So nothing to worry about here.

Then I was reading other forums such as the VFX subredit and a lot of people are worried and agreeing that the field is dying a slow death. New people who studied VFX or similar are looking for career change. While veterans are seen it coming and take solace that they are not far away from retirement age.

Am I reading too much into this? Is the recent progress in AI just the next Bitcoin/IPv6/3D TV's..etc? or do we need to prepare for the worse right now?

[rossdavidh]

One thing I recommend, to gain perspective on this, is to replace "AI" with "LLM" (large language models) in any sentence about the likely impact. For example, instead of "the impact of AI on developers on HN", think "the impact of LLM on developers on HN". Does it seem like large language models will replace developers?

The comparison to bitcoin is apt, in that the blockchain (not quite the same thing but related and part of the same hype cycle) is actually a useful algorithm, for certain applications. Way, way fewer applications than was believed at the peak of the hype cycle.

The thing is, how one prepares for wider use of LLM's, would be to make sure you're doing the things that LLM's are bad at, which is:

1) being honest

2) knowing the fundamentals of how the system you're programming for works, instead of just regurgitating some code you found elsewhere

3) try to make the simplest code that gets the job done, rather than find an excuse to use the latest buzzword to make things complex

This is all not only what would help to distinguish you from what an LLM would produce, it is also all good stuff to do regardless of whether or not LLM's flame out like many hype cycles before, and we enter a third "AI winter".

By the way, if you're not familiar with the term "AI winter", now would be a good time to look it up.

[weatherlite]

> Does it seem like large language models will replace developers?

To me it looks very plausible they will reduce the amount of programmers needed at the very least.

[ungreased0675]

There have been several attempts in the past to replace programmers, from easy to use languages to no-code drag and drop tools. I believe LLMs are a little different than past attempts, but I’d hesitate to say how different.

Over the past few weeks I’ve used Copilot to teach me how to create some fairly complex process automations. That probably took a job away from a programmer, or maybe the task wouldn’t have gotten done at all. I’m not sure.

[weatherlite]

I think the job is going to change tremendously soon, you will still need to be technical but you'll spend less time coding/looking at code and more time prompting the A.I to do it for you. On the long run for sure the job will go away, as all jobs will. No one knows how fast this will happen, could be 7 years and could be 27 years.

[transfire]

From what I have observed the current generation of AI (LLMs) is only good enough to help a programmer — write example algorithms, give ideas for what direction yo take, etc.

However, we are about one major breakthrough away from total game change. How long before that happens?

[h2odragon]

In addition to the usual hype cycle bullshit; "AI" is faced with the problem of being heralded as something between an infallible oracle and an acceptable God. Potentiates the bullshit and draws in participants from further afield than the usual "VR, this time fer sure!" parties have.

The previous incarnations of AI hype have been so oversold as to lead to the previous "AI winters." There's real progress this time too, but it's incredibly hard to reason about or criticize because its so heavily covered in marketing hype and scammer scrap. After a few years the real advances will crawl out into the world and be seen for the incremental, useful tools they are.

So, speaking for myself; I'm not in denial; I have hopes: I'm just cynical about the eventual realities and deeply cynical about the time-frames in which they will develop.

[silverquiet]

I don’t see how a sometimes-demented chatbot is going to revolutionize anything. That’s not to say I’m not impressed with what they can do these days.

I’m also impressed that a car can mostly drive itself these days and only rarely veer into oncoming traffic, and yet humans still drive the vast majority of all vehicles.

Basically, I’ll believe it when I see it.

[gregjor]

HN doesn't really qualify as a "community." It's site full of mostly tech-related or tech-adjacent posts and comments, viewed and added to by people from all over the world, of all ages and experience levels, interests, motivations, etc. So it doesn't make a lot of sense to look for a consensus or overall opinion, or to ask if "HN is in denial" because HN doesn't stand for any coherent group of people or set of opinions.

I have seen HN opinions about what we're calling AI this year range from fear of the robot overlords, to wide-eyed acceptance of the hype and reverence for Sam Altman, to wait and see, to skepticism and outright dismissal of the whole thing as another grift run by VCs. So opinions range all over the place, as does the reasoning expressed to support them.

No one knows what will happen, other than the obvious: Lots of money will get poured into AI, and the technology will get deployed (almost certainly prematurely because of FOMO), and pushed on us whether we want it or not. That's already happening. Whether the money and the hype will actually accomplish the stated goals of either AGI or vastly improved productivity, who knows. So far we don't even have full self-driving or LLMs that don't just make things up.

Some people, like Sam Altman, seem to think AGI or at least productive uses superior to humans comes down to scaling LLMs up. Other people think that LLMs will plateau, reaching a limit of capabilities inherent in how they work and increasingly polluted and "photocopy of a photocopy" training data as future LLMs train on LLM-generated content of questionable quality. Follow the money and that very likely points to who stands to profit from LLM hype and adoption.

I've worked in IT as a developer and system admin for over four decades, so I remember plenty of previous hype cycles, including the second "AI winter" (the first AI winter happened when I was still in high school). And I've survived multiple apparent threats to my career future, mainly offshore outsourcing, but also things like no-code/low-code, trends that stoked a lot of worry but either fizzled out or caused more of a ripple than a tsunami. I agree with the people who have pointed out, on HN and elsewhere, that writing software requires a lot more than learning a programming language and cobbling together snippets gleaned online, though a lot of people get jobs in the industry with those skills and little else (or used to).

Some specific fields look more vulnerable than others. News and "content production" have already taken a hit. I wouldn't tell my kids to go into transcription, translation, paralegal, or even law at this point. Maybe avoid rote medical jobs like looking for anomalies in lab and imaging results. VFX looks vulnerable, but that field got computerized a long time ago. For everyone amazed at Sora's videos (which I admit seem impressive), consider we've had Pixar-style movies and photorealistic video games for quite a while without anything we might call AI. We could experience significant hollowing out of some industries, very similar to what happened to auto manufacturing, and manufacturing in general, due to both automation and offshoring.

Because I write code for a living I like to think that won't get taken over by LLMs in my lifetime. I've looked at the current crop of tools, I'm not worried, but junior developers might consider the impact of those tools on their prospects. I will retire when I can't buy a keyboard that doesn't have a copilot key on it.

[hdlothia]

Interesting, why not law? It feels like it would be one of the last fields to be automated away. Especially trial lawyers.

[gregjor]

Most lawyers don't do trial work. They put together boilerplate contracts, read through discovery documents and highlight and summarize, pretty rote stuff. Trial lawyers probably don't have to worry for now, but a lawyer specializing in, say, filing bankruptcies (already almost completely automated without LLMs), or filing disability claims, or drafting contracts, leases, etc. -- the vast majority of the output of the legal profession -- may get run over by LLMs.

Nolo Press kind of got there decades ago, with books about legal procedures and tear-out fill-in-the-blanks forms sold in bookstores. No LLM needed, it made the law accessible to a lot of people. Plenty of areas of routine law get handled by automation or outsourcing to non-lawyers already. Think about a residential lease, basically a word processing template with the blanks filled in by software.

Look at tax preparation to see where at least some large piece of legal work will probably go. A vanishingly small number of people need a tax attorney. Most people can use software, or go to an H&R Block office or pay an accountant to prepare their taxes, with "pro" versions of the same software. Similar software already exists for things like bankruptcy, divorce, contracts, residential and commercial leases, on and on. The legal profession is already heavily reliant on computer software. Whether LLMs will make that software better or not, I don't know.