Sam Altman:
A lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we’re going to be able to run them on our computers, and so we’re not going to need all the stuff in the data center.
But from a global allocation perspective, actually it’s quite terrible to have everybody, like, have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used five percent of the time.
Like you want maximum utilisation in the cloud if you’re trying to deal with this very supply-constrained environment,
And somehow many people’s intuition is the opposite of that
TL;DR Guy selling product: “Here’s why owning your own product instead of renting it from me is a terrible idea”
See, the thing is though, all of these things are designed with some kind of running hours in mind. If a car is used more, it’ll wear our faster so it’s either using it a lot in a short time or less in a long time, its utility ends up being the same so there’s no direct gain from not letting it sit idle. That’s why passive robotaxis are a pretty dumb idea.
But yes there are things that really only wear out from time spent existing. A pencil’s lifespan is purely defined from its use, a cast iron pan can pretty much be used indefinitely. Cars and drills are more like pencils imo. RAM might be more like a cast iron pan though, although trace erosion is still a thing.