Hacker News Reader

Comment by jeanlucas | Hacker News Reader

So convenient a future AI dev will cost as much as a human developer, pure coincidence

Similar to housing in attractive places no? Price is related to what people can afford, rather than what the actual house/unit is worth in terms of material and labor.

Except for "material and labor" there is an additional cost of land.

That is already "related to what people can afford", in attractive places or not.

It's sort of like how high cost funds net of fees offer the same returns as low cost ETFs net of fees.

I'm not sure I understand this one.

Even with the downvote, I still don't understand the logic of the original comment...

High cost funds that will take your money perhaps (top funds are almost all capacity constrained, but will deliver a Sharpe of about 2 or 3)

It would be interesting to study the returns of funds not taking new investors and funds taking new investors that are run by the same manager. I bet there's quite a performance difference. One giant of the industry comes to mind.

This is just a ball park number. Its like the AI dev will cost some what less than a human developer. Enough for AI providers to have huge margins and allow for CTOs to say - "I replaced all devs and saved so much money".

And then the CTOs will learn the truth that most product managers are just glorified admin assistants who couldn't write a spec for tic-tac-toe.

And that to write the business analysis that the AI can actually turn into working code requires senior developers.

The full cost of an employee is a fair bit more than just their base salary.

Wait until the taxes on AI come, to pay for all the unemployment they are creating.

I mean, hey, rather than use AI at work, I'll just take the extra $100k/year and be just that good.

Cost as US developer or Indian developer?

But he works 24/7 at then maybe 20x output

Like fuck that's happening. Human dev will spend entire day gaslighting an electronic moron rather than an an outsourced team.

The only argument we have so far is wild extrapolation and faith. The burden of proof is on the proclaimer.

Why we can’t keep the current jobs but accelerate humanity development by more than 20x with AI? Everyone is just talking about replacement, without the mention of potential.

I'm not entirely sure I understand exactly what you're suggesting. But I'd imagine it's because a company that doesn't have to pay people will out compete the company that does.

There could be some scenario where it is advantageous to have humans working with AI. But if that isn't how reality plays out then companies won't be able to afford to pay people.

I don't think there is market demand for 20x more software produced each year. I suspect AI will actively _decrease_ demand for several major sectors of software development, as LLMs take over roles that were handled previously be independent applications.

Right. This is insightful. It's not so much about replacing developers, per se. It's about replacing applications that developers were previously employed to create/maintain.

We talk about AI replacing a workforce, but your observation that it's more about replacing applications is spot on. That's definitely going to be the trend, especially for traditional back-office processing.

I'm specifically commenting on the double whammy of increased software developer productivity and decreased demand for independent applications.

I think it depends on how you view it. With 20x productivity you can start to minimize your supply chain and reduce costs in the long term. No more cloud usage in foreign countries, since you might be able to make the necessary software by yourself. You can start dropping expensive SaaS and you make enough for your own internal needs. Heck, I would just increase the demand because there is so much potential. Consultants and third-party software houses will likely decrease. unless they are even more efficient.

LLMs act as interfaces to applications which you are capable to build yourself and run your own hardware, since you are much more capable.

> and third-party software houses will likely decrease. unless they are even more efficient.

It's going to be really fun for us people who love to write unicode symbols into numeric input boxes and such funny things.

That's not how this works:

https://en.m.wikipedia.org/wiki/Jevons_paradox

Jevon's Paradox isn't a hard rule. There's plenty of situations where a resource becomes cheaper and the overall market decreases in total value.

The odds of automation not being like energy are super low. Aka the more, the better.

There is great potential. But if humanity can't share a loaf of bread with the needy, nor stop the blood irrigation of the cracked, dusty soil of cursed Canaan[^1], what are the odds that that acceleration will benefit anybody?

([^1]: They have been at it for a long while now, a few thousand years?)

And is neither reliable nor liable.

An LLM by itself has 0% output.

An engineer shackled to an LLM has about 80% output.

Reply on news.ycombinator.com