undefined | Better HN

0 pointsautomatic61311y ago0 comments

You're taking exponential improvements for granted. Only one human endeavour (ever) has made exponential gains for a long period, that of silicon lithography. And even that ended in the last decade.

The reason I say "always underestimate" is the obvious one. The easy problems are solved first, then the harder ones take longer because they're harder.

So that's fantastic that battery production scaled up. Only a fool would expect - would plan on - it continuing like that.

0 comments

huijzer1y ago

That’s not true. Exponentials are everywhere. They can often go for a long time. Even silicon is still going strong if you focus on FLOPS per dollar.

Other examples are cruise ship sizes, US GDP per capita, or Microsoft stock price.

The point really is this: go back in time to some of these things a few decades years ago. You would say: “How is this even possible? This is crazy. This will probably plateau soon. This can’t continue.”

But it did. Cruise ships went from 20, to 100, to 200 and now 365 meters in length.

And the same will probably happen for batteries. People say “ah well this is crazy. It will probably plateau soon”. My point is maybe it won’t. Once these exponentials (starts of s-curves) go, they go. Standing at the bottom of an s-curve and predicting the plateau soon can lead to a massive misprediction. Like the IBM CEO who said there will never be a market for more than 10 computers. He was off by about a billion.

automatic6131OP1y ago

Fyi, monetary values of things, like US GDP or stock prices, can be exponential forever, if we wanted, because they're socially constructed.

I am talking about real things. Cruise ship sizes improved dramatically but... actually linearly? There won't be 1,000,000 GT cruise ships in 2050. Or 60.

You have to use specific measurements like FLOPS/$ to keep moore's law alive, because the focus has been only on a certain kind of FLOP (the fp32 MAC for graphics, or perhaps the INT8/FP8 in recent years). Because in general, it's dead. It's more performance, for more money. Because lithography is really hard, harder now than ever (and water is wet).

j / k navigate · click thread line to collapse