Comment by simion314 | Hacker News Reader

Comment by simion314 | Hacker News Reader

Parent Full thread

Assuming LLM improve productivity why do you think the market will shrink instead of getting even more stuff done? Say if some actually smart LLM could improve my team productivity with 25% then

1 we could fire a person

2 we could handle 25% more tasks, features and bugs

So I think 2 will happen.

My latest example of LLM issues,| I asked ChatGPT and competition to convert a line of jQuery into native JS, they all got it wrong because jQuery has some selectors like ":header" that is native to JS but the LLM used it anyway though somewhere in it's big memory it has the information that his is not native and if you prompt it right it will fix the issue. So seems to me the LLM are focusing too much on the prompt and failing to use it's full memory on the problem, so it can fix small individual micro tasks but is a complete waste of time for something a bit more advanced , my conclusion you can't have a manager or an artist armed with ChatGPT and create a full project without actually learning to code, at best they will learn to code from the LLM but the hard way and probably using outdated code and inefficient ways.

>Assuming LLM improve productivity why do you think the market will shrink instead of getting even more stuff done?

Because this is what always happens when tools cause productivity gains. Budgets aren't infinite.

- Computerized avionics dramatically cut demand for flight crews. It used to require a dedicated "flight engineer" to literally run the engines (like a train engineer) in a WWII-era bomber, and a dedicated navigator. Now airliners barely need a co-pilot.

- As I mentioned previously, computerized medical imaging has caused demand for radiologists to go down, because one radiologist today can do the work of multiple radiologists of yesteryear.

- In software in particular, if engineering and development is more productive, companies can cut costs and still get the same amount of work done, or possibly more. It's not about firing people immediately, it's about the next company hiring fewer. And the one after that hiring even fewer. Generative coding is going to make software development labs leaner.

>my conclusion you can't have a manager or an artist armed with ChatGPT and create a full project without actually learning to code, at best they will learn to code from the LLM but the hard way and probably using outdated code and inefficient ways.

That's the current state of affairs. Why do you think the technology will not improve beyond where it is now?

> As I mentioned previously, computerized medical imaging has caused demand for radiologists to go down

And as I said in my reply, this isn't true.

Maybe I was wrong on that point, and maybe I wasn't. Are you suggesting it invalidates the entire premise? Or is this just classic HN pedantry?

Do you have anything useful to add to the discussion?

I don't care about the rest of the post, or about arguing with you. I'm publicly correcting a specific, widely repeated, false claim. One which, if you take the time to read the links I posted, you will see is having a concrete negative effect on an industry responsible for real people's wellbeing. In particular (see the third link), ~50% of students considering radiology as a specialisation express concerns because of false claims its future as a job is under threat. These concerns contribute to hiring problems in that profession, which exacerbate a global staff shortage (first link) which results in worsened healthcare outcomes.

There's a non-zero chance that correcting your statement will influence somebody considering whether to specialise in this field, and replying to your repeating of the claim as well as to the original instance increases that chance. I don't consider it pedantry.

You're seriously trying to claim being factually right doesn't matter in a discussion of facts?

I think the misconception is that there's a finite amount of stuff to do. Every place i've worked at there's been a near infinite backlog. Lots of stuff that would be done if they could. It's always been a constraint on the number of people hired (budget and availability).

I don't see engineers not being part of the process for many years, even if most of the work is done for them. For most companies, if they had their engineers suddenly become 10x more productive, they'd want them to do 10x as much, not scale down to 10% of the staff. Not to say some companies wouldn't.

There are certainly more programmers now than when you had to build a program by wiring together vacuum tubes, or even manually manage memory. Hollywood definitely didn't employ 2.5 million people back in the silent film era when film-making tools were much less productive. That's Jevon's Paradox, increased efficiency increases demand. Labor is like 30% of an airline ticket, and that counts not just the flight crew, so efficiency doesn't do much much for demand when jet fuel is pricey. Likewise there's only so many Xrays that are justified to give out. There's an infinite amount of software crud to be created, and it's primary input is programmers.

>That's the current state of affairs. Why do you think the technology will not improve beyond where it is now?

To replace a developer you would need a human level AI, that is impossible for now.

You can improve the tech to do simple jobs and help me like

- review this code I wrote for potential corner cases

- generate some unit tests for this functions

- find some code in this giant project that does X . I know I wrote it 3 years ago but I have no idea how I named the function or maybe it was a different project.

- rewrite this old IE6 JS/jQuery into modern JAvascript

But IMO a tool that can do this would use LLM but you need an actual valid code checker/interpretor to do a production ready code. I can't trust an LLM to update a giant piece of code without chainging something, I prefer a tool like Intellj refactoring that actual understand code.

>Because this is what always happens when tools cause productivity gains. Budgets aren't infinite.

I think you'd actually struggle to find an example in history where productivity gains have resulted in prolonged, and industry-wide unemployment for those trained in that area.

I completely agree that technology can make individuals redundant, as in your flight crew example. And yet we are seeing a global shortage of airline pilots, how do you explain that? If technology makes people unemployed (across an economy, over the medium-long term), why aren't there piles of pilots working at McDonald's?

Is not that as technology improves, industries which were once small/unprofitable can balloon into hugely lucrative industries?

Reply on news.ycombinator.com