Hacker News Reader

Comment by lithocarpus | Hacker News Reader

I don't see how over time we could get to a place where an entity with orders of magnitude less computing power can run AI that is anywhere near as powerful as the huge companies. Maybe for certain narrow applications, maybe even for many such applications, but hard to imagine it happening in a way that un-concentrates power.

Though certain novel uses could lead to new individuals or entities gaining power.

I'd like to be hopeful and would like to hear good arguments for how this could happen - but it seems to me improved technology on the whole leads to increased concentration of power - with exceptions and anomalies but that being the dominant trend.

View on HN

I am sure the same thing was thought about never getting to a point where every person would have a tiny computer in their pocket that was orders of magnitudes faster than the multi-million dollar computers that took up whole rooms and were only owned by the largest companies.

You mean a tiny computer in your pocket that you can mostly just consume content with and chat with others. Sure there are some neat, useful apps out there, but you can't really learn how technology works on a tablet or phone unless you're allowed to build that specific kind of app. You don't even have a filesystem. 10 year old me in today's world would have not been able to break my phone like I did my old DOS / Win95 PC and actually learn something. Shit I used to spend hours just browsing the filesystem, and the install media to see what I could find and learn and use. That's how I found the Weezer music video, and the pinball game on the Win95 I still CD. There is no equivalent to this with phones.

My comment has nothing to do with that.

It was about how only big companies have the resources to make big computers that take up a whole room that are powerful enough to run smart AI models.

But if tech progress is any indication, in say 50 year or probably less, we will have the power of a modern day datacenter in our pockets and be able to run smart AI models locally without it being a large corp monopoly.

If you take human brains as the limit (a questionable assumption) they do it all for 20 watts (and hardware that makes itself!) The training is often years and can be expensive.

In all seriousness though there’s plenty of room for improvement both in current models and hardware.

(Apologies for the bad form of replying twice…)

> it seems to me improved technology on the whole leads to increased concentration of power

Which is why we are dominated by IBM, AT&T, Kodak and Xerox.

Reply on news.ycombinator.com