Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Jevons paradox

en.wikipedia.org

Post reply on HN
Print view View original

61–70 of 165 posts

Re: Jevons paradox

#61

Post by nine_k »

harrison_clarke wrote: post #41
TSiege wrote: post #16

Earlier quoted context omitted.

Well I can answer one question for you. There are studies that show that better insulation only briefly lowers the energy use. Once people adjust to the new insulation they end up heating or cooling their houses more since in their mind it is cheaper, but they end up using the same or more energy. https://www.cam.ac.uk/research/news/insulation-only-provides...

presumably there are limits to that. once every room is the preferred temperature, i'm going to stop running the heater although maybe i'd get bored and run the heater and the AC at the same time

Often you can't because the heater and the AC is literally the same unit, the heat pump.

Re: Jevons paradox

#62

Post by whatshisface »

cjbgkagh wrote: post #46
aurareturn wrote: post #39

Earlier quoted context omitted.

new method was 1,000,000x more efficient would there still be more total money spent on GPUs because of it In my opinion, yes. If scaling laws continue at 1,000,000x, we will turn the whole planet (or Mercury) into a giant GPU.

Do you have a use case for your opinion? On the assumption that we’d want super intelligence to do interesting things, consider that we currently have many very smart humans and society is not maximizing their use, quite the opposite.

Very smart humans don't want to do random tasks, they want to study number theory.

Re: Jevons paradox

#63

Post by cjbgkagh »

roenxi wrote: post #44
cjbgkagh wrote: post #36

The dose makes the poison, what if a new method was 1,000,000x more efficient would there still be more total money spent on GPUs because of it. What if efficiency brought it down such that inference on CPUs that people already have is good enough. I see a large demand curve but not an infinite one. For example laptops are getting faster every year but most people I know are not clamoring for the latest laptop, same…

> what if a new method was 1,000,000x more efficient would there still be more total money spent on GPUs because of it Obviously yes; 6 orders of magnitude would result in insane amounts of money swooping in. The market has currently been going crazy over relatively minor improvements compared to that. > If a laptop came out that was 50x faster for the same price people wouldn’t buy 50x more laptops, they would wait…

Hence my very first point; the dose makes the poison.

Yeah, it would be super impressive, perhaps another point SMT solvers algorithms have improved faster than the hardware for a very long time and are incredibly useful. But we are not spending infinite money on SMT solvers, quite the opposite.

Re: Jevons paradox

#64

Post by steveBK123 »

Given the likes of sama & co were asking for unfathomable levels of capital, chipmakers had backlogs, datacenter & power supply chain companies were getting bid to the moon, etc.. having a dramatic decrease in compute requirements might just bring things closer to a supply/demand balance.

Where was Stargate gonna get $500B? How was Altman ever going to raise.. $7T? (~1/3 of the entire GDP of US)

Does a 10x or even 40x decrease in compute cost just mean they can actually raise & spend the capital that their compute requires?

Re: Jevons paradox

#65

Post by belter »

ijidak wrote: post #45
llm_trw wrote: post #8

Earlier quoted context omitted.

Having read the paper from deepseek I seriously don't get how anyone can think "Oh now is the time to sell nvidia". Both at the tactical level and strategic level that paper means we will see _more_ nvidia gpus fly off the shelves. If anything it's showing why the h800 is a decent card and everyone in China should buy one (dozen thousands).

But, I thought the CUDA moat is primarily needed for training. On the inference side, why does one need Nvidia GPUs over Intel and AMD GPUs?

Tensor Cores accelerate inference...

Re: Jevons paradox

#66

Post by cjbgkagh »

whatshisface wrote: post #62
cjbgkagh wrote: post #46

Earlier quoted context omitted.

Do you have a use case for your opinion? On the assumption that we’d want super intelligence to do interesting things, consider that we currently have many very smart humans and society is not maximizing their use, quite the opposite.

Very smart humans don't want to do random tasks, they want to study number theory.

Not all, and many smart humans are prevented from doing what they want to do for all sorts of reasons. One reason being others feel threatened by them, but those same people will feel threatened by AI, probably more so.

Re: Jevons paradox

#67

Post by roenxi »

cjbgkagh wrote: post #63
roenxi wrote: post #44

Earlier quoted context omitted.

> what if a new method was 1,000,000x more efficient would there still be more total money spent on GPUs because of it Obviously yes; 6 orders of magnitude would result in insane amounts of money swooping in. The market has currently been going crazy over relatively minor improvements compared to that. > If a laptop came out that was 50x faster for the same price people wouldn’t buy 50x more laptops, they would wait…

Hence my very first point; the dose makes the poison. Yeah, it would be super impressive, perhaps another point SMT solvers algorithms have improved faster than the hardware for a very long time and are incredibly useful. But we are not spending infinite money on SMT solvers, quite the opposite.

But we are spending more than ever before. I don't see why you think being unable to spend infinity dollars matters here. There are a lot of numbers between larger than current spend between where we are and infinity.

Re: Jevons paradox

#68

Post by cjbgkagh »

roenxi wrote: post #67
cjbgkagh wrote: post #63

Earlier quoted context omitted.

Hence my very first point; the dose makes the poison. Yeah, it would be super impressive, perhaps another point SMT solvers algorithms have improved faster than the hardware for a very long time and are incredibly useful. But we are not spending infinite money on SMT solvers, quite the opposite.

But we are spending more than ever before. I don't see why you think being unable to spend infinity dollars matters here. There are a lot of numbers between larger than current spend between where we are and infinity.

To my point, how much do you think a SMT solution costs now?

Re: Jevons paradox

#69

Post by YurgenJurgensen »

marcosdumay wrote: post #57
tbriudmepn wrote: post #3

Jevons Paradox Paradox: Markets fall on news that should make them rise, due to Jevons Paradox.

Somehow, nobody expects the Jevons' Paradox. Even when those same people talk about it every other day.

The Hofstadter-Jevon Corollary: Taking Jevon’s Paradox into account only makes it worse.

Re: Jevons paradox

#70

Post by roenxi »

cjbgkagh wrote: post #68
roenxi wrote: post #67

Earlier quoted context omitted.

But we are spending more than ever before. I don't see why you think being unable to spend infinity dollars matters here. There are a lot of numbers between larger than current spend between where we are and infinity.

To my point, how much do you think a SMT solution costs now?

Couple of thousand dollars? Consultant turns up on site, gathers data, talks to a few people, sets up a model and solves it. Probably around a weeks work @ say $100/hr. Might be under-calling it for validation and there'll probably be some iteration to do so a project will take a few solves to complete + overheads.
Post reply on HN