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

31–40 of 165 posts

Re: Jevons paradox

#31

Post by spamizbad »

llm_trw wrote: post #24
spamizbad wrote: post #20

Earlier quoted context omitted.

This is just a baseless conspiracy theory that I have, but I do wonder if they intentionally avoided certain avenues of research because it could erode the “moat” major tech firms have by dramatically reducing the capital costs required for training and inference. If you focus your research around work that mandates high-end hardware at scale you can lock out tons of potential incumbents

There is nothing in the deepseek paper that suggests you can't use the order of magnitude in hardware costs you saved to just train models that are ten times as large.

But there are thresholds of commercial viability in all of this. DeepSeek's technology gets you over that line with less sophisticated hardware.

There's already some pretty impressive work being done with folks using just a pair of M2 Ultras with r1 in a "home lab" context that goes way beyond what you could previously do with llama.

Re: Jevons paradox

#32

Post by pjc50 »

mrweasel wrote: post #12

This has actually been on my mind lately and I've been looking for statistics, from Denmark, but haven't been able to find any going back far enough. My assumptions is that we in the late 1940s through the 1950s had achieved a reasonable standard of living, but perhaps not available to all. Even if we assume that everyone in a country has access to the same, high, late 1950s standard of living, but applied modern tec…

> same, high, late 1950s standard of living

.. I suspect this is wildly less good than most people would expect, along with forgetting about how widely distributed it was or wasn't.

Just look at the first chart here: https://www.bbc.co.uk/news/uk-42182497

If you could get people back to 1950s level of travel, whether in 1950s vehicles or (way better!) modern EVs, that would make a huge difference to the environmental impact. But of course if you actually suggest measures to force that people (correctly) notice it would reduce their standard of living.

Re: Jevons paradox

#33

Post by mminer237 »

casey2 wrote: post #22

The intuition behind this is that people who had a usecase for the tech but were locked out because they couldn't afford it, or didn't believe it was worth it usecase can be engineers or researchers. Some of these people may perform better than expected causing a gap in true market value which is filled when they scale up to their new size Related: Jenson's Paradox: The more you buy, The more you save

I guess my question about the current events is: Are people really locked out of LLMs due to price? It seems like everything already has AI in it and that virtually every end user hates it. I could see a shift to more local models or something, but not an increase. I feel like LLMs have largely been oversold as a solution in search of a problem.

Or are people just applying this as everyone and their mothers are going to start training competing models to carve out their piece of the market?

Re: Jevons paradox

#34

Post by HeatrayEnjoyer »

StefanBatory wrote: post #30

I see it already mentioned in the page ;) Jevons Paradox and DeepSeek AI The Jevons Paradox, an economic principle stating that increases in efficiency often lead to higher overall consumption, has been observed in the context of DeepSeek AI. DeepSeek, a Chinese AI startup, recently introduced its R1 model, which achieves comparable performance to leading AI systems like OpenAI's ChatGPT while requiring significantly…

Deepseek is 651B, GPT models are suspected to be <200B. Where are they getting the idea that Deepseek is more efficient?

Re: Jevons paradox

#35

Post by throwmeme888 »

strobeflier wrote: post #28
mrbungie wrote: post #27

Earlier quoted context omitted.

Sure, 31 December 2024 is by definition last year, but at the same time that was not even a month ago.

Come on. Which do you think is more likely, that it was a normal EOY sale of a lucrative stock, or that they sold based on insider info obtained weeks before anyone else?

it was 2-3 days after deepseek paper release on Christmas Day. I read this paper too and drew the exact same conclusion (that nvda was due to decline)

but I dont think its insider trading, just informed and reactive trading - not eoy profit trading either though

Re: Jevons paradox

#36

Post by cjbgkagh »

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 with phones, good enough is good enough and people would rather save the money. If a laptop came out that was 50x faster for the same price people wouldn’t buy 50x more laptops, they would wait even longer to upgrade.

Re: Jevons paradox

#38

Post by sz4kerto »

mminer237 wrote: post #33
casey2 wrote: post #22

The intuition behind this is that people who had a usecase for the tech but were locked out because they couldn't afford it, or didn't believe it was worth it usecase can be engineers or researchers. Some of these people may perform better than expected causing a gap in true market value which is filled when they scale up to their new size Related: Jenson's Paradox: The more you buy, The more you save

I guess my question about the current events is: Are people really locked out of LLMs due to price? It seems like everything already has AI in it and that virtually every end user hates it. I could see a shift to more local models or something, but not an increase. I feel like LLMs have largely been oversold as a solution in search of a problem. Or are people just applying this as everyone and their mothers are going…

> Are people really locked out of LLMs due to price?

Yes. For example Google has just made Gemini a standard part of Workspace, before that it cost $36 or something per month per user. That was too much for many SMBs to experiment with (you and I understand that the potential efficiency gains are way higher but for an SMB, paying 4x more for your office suite sounds bad).

Re: Jevons paradox

#39

Post by aurareturn »

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…

  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.

Re: Jevons paradox

#40

Post by mrbungie »

strobeflier wrote: post #28
mrbungie wrote: post #27

Earlier quoted context omitted.

Sure, 31 December 2024 is by definition last year, but at the same time that was not even a month ago.

Come on. Which do you think is more likely, that it was a normal EOY sale of a lucrative stock, or that they sold based on insider info obtained weeks before anyone else?

Parent never suggested it was due to DeepSeek specifically, but for supossedly structural reasons if you follow the first paragraph.

I'm just calling out that "last year" wasn't really "last year" if you follow that argument.

Post reply on HN