Hacker News Reader

Comment by rwyinuse | Hacker News Reader

I agree. Programming was just a natural first problem to tackle, but the impact will not be limited there. If you give today's best model access to all the data sources and systems an average white collar worker has, it probably could do most of their job. But organizational change takes its time, systems need to be integrated into new AI tools and so on, so it will take a while.

People tend to overestimate change in short term, and underestimate it in long term. AI hasn't replaced programmers yet, but I believe it will replace most white collar workers in 20 years.

View on HN

You actually claimed two very different things.

1) that LLMs can do existing white collar work well once some (possibly significant amount of) work is done to fully plug them in and get specifications settled.

2) That all white collar workers will be replaced by LLMs.

Perhaps a typo, but I don't think (1) implies (2).

> and get specifications settled.

In my ~20 year career in software, I've never once gotten adequate specifications. I honestly don't feel threatened.

> In my ~20 year career in software, I've never once gotten adequate specifications.

I haven't either (in my ~30 year career) other than maybe one or two rare occasions, but I also believe that having adequate specifications matters much less for the vast majority of software when the time & cost to fix a gap after you notice it shrinks to nearly nothing.

In the past bad specs resulted in huge problems due to process log-jams that came out of having to coordinate multiple human workers around the spec. But increasingly the person who poorly specified the thing the first time around can rectify the problem with a couple of new prompts.

Indeed, however thanks to the improvements in automation, from programming languages to cloud environments, projects that used to be done with a team of 20 persons, nowadays can deliver the same outcome with a team of 5.

It sucks to be one of those 15 left out.

Only if the number of projects stays fixed. Those 15 can form 3 new projects.

Only if we follow the MBA curve of exponential growth, which is physically impossible, because there aren't only 15 on their own, rather multiplied by amount of existing projects dropping devs on bench, per companies targeting the same industry sectors, and then wiped out by the amount of customers willing to pay for delivery projects.

The "there is work for everyone" only works, when we ignore offer and demand.

Works great on VC sales pitches, though.

I dunno, I look around and I see a lot of work that needs to be done, and I see a lot of people locked up doing useless things. The work isn't in the same industries that are optimizing. For example, local construction companies are now spinning up dev teams where they wouldn't before, because local devs are more plentiful now they're not flocking to silicon valley.

So right as silicon valley is optimizing themselves out of devs, the rest of the world is optimizing themselves out of needing silicon valley. This one construction company I know about had just spent at least $300k on Oracle services, they will now be redirecting that spend to their in-house team using AI. Whether there are enough opportunities for this in the long term, I dunno, but I sure hope it means people will be working on problems other than optimizing ad placements or extracting consulting fees.

I think the question of whether (1) implies (2) is the core of AI economic concerns. Will this be another 'normal' technological revolution where the jobs and workers adapt to provide value on top of the models that are able to automate their current work, or will there simply not be room in many areas to provide additional value on top of the automated work?

I think the latter would not occur under the current state of AI/LLMs, but the rapid pace of growth gives me pause.

I go back and forth. You can go mad reading the tea leaves.

Reply on news.ycombinator.com