The author makes an error while explaining Jevons’ Paradox. They say “the cheaper and faster we got at producing coal, the more coal we ended up using”. But that’s not paradoxical at all. That’s basic supply and demand curve from economics 101. It’s exactly what you’d expect.
Whereas if you take the extraordinarily difficult step of opening Wikipedia (https://en.wikipedia.org/wiki/Jevons_paradox), you understand the real paradox. Jevons observed that when steam engines get more efficient and individually use less coal, we end up deploying many more steam engines in many new industries and use cases, increasing coal consumption overall.
This seems like a basic oversight, maybe this guy hasn’t heard of Wikipedia. On the other hand, I think he wants to push the narrative that AI is seeing enormous productivity gains. He does this by using price per token, which has fallen. Very similar to Jevons supposedly observing “coal production”. But the author is struggling to point to new industries and use cases that were opened up by AI, like when we deployed steam engines back in the 1860s. So he misstates Jevons paradox, removes the paradoxical part, and makes it seem like his thesis makes sense.
Jevons’ Paradox could still apply here! I’m not saying it doesn’t. But we just haven’t seen the examples quite yet. A good example would be an observed surge in demand custom built software as software engineers become more efficient. But lower token costs ain’t it.
> price per token, which has fallen.
I think that statement is incomplete. It's cheaper because the AI providers are subsidizing the queries by burning cash in order to gain market share. So it's not a price, it's a subsidized and temporary price. Which will likely go up once burnable money runs out / the providers switch from "market acquisition" to "let's try to make this thing profitable at all".
Also, I'm sure I read something here in the last few weeks documenting that while every specific model shows falling price-per-token, "frontier models" which are the ones everybody wants to use - maintain steady prices (and are often significantly more expensive to use because they consume way more tokens than older models responding to prompt-fashion-de-jour).
I'm not sure this is true anymore. Anthropic e.g. claims that each model is profitable, seen over its lifetime. There are also a ton of providers of open-source models that are popping up that are hosting models for very cheap prices per token - hard to say how subsidised they are, of course, but I would be surprised if they are all burning cash.
Open-source models are also constantly getting better, with new architectures and training processes that make 30b models perform like previous SOTA models, lowering the prices for many use-cases.
So I don't think the current level AI will get more expensive, I would expect it to stay at the same price or fall. But, I think the AI companies will continue pushing more higher-priced premium models that deliver better performance for higher prices and with more tokens consumed.
This is already happening. Some time ago all models providers charged $20 a month. Now you still get $20 version, which is crippled, with limits. If you want "a real thing", you need to pay $200. And this might still be too little, as the algo that runs the show is quadratic with the size, which, as every algos books says, is highly inefficient.
[dead]
a16z is heavily long on AI, so this article sounds very biased.
From the article: If you live in the United States today, and you accidentally knock a hole in your wall, it’s probably cheaper to buy a flatscreen TV and stick it in front of the hole, compared to hiring a handyman to fix your drywall.
Probably because the US has been focused on services for years rather than physical goods production. Everything else in US is focused on importing cheap(er) goods or materials.
> On the other hand, I think he wants to push the narrative that AI is seeing enormous productivity gains.
That is my impression as well. I would be thrilled to see this mythical 10x productivity. Even with 2x productivity, I would be highly pleased. This should mean developers (and everyone else) are producing 2x more quality, software (and general services) are 2x better? I see none of that, except 2x more junk. Did AWS, GCP, or anything else become 2x cheaper and 2x more stable? Maybe I'm living under a rock.
Also the initial claim is just false — “if you live in the United States today, and you accidentally knock a hole in your wall, it’s probably cheaper to buy a flatscreen TV and stick it in front of the hole, compared to hiring a handyman to fix your drywall“. I can tell you that this isn’t true in San Diego. Unless they’re using flatscreen tvs that cost less than $300? Or perhaps making extremely difficult-to-patch holes somehow
I live in the UK, and it’s basically £100 to get a tradesperson to show up to my front door regardless of what I want them to do. I can buy a flat screen TV from £100 new from one of the UK high street retailers [1] , or £85 [2] if I go on Amazon
[1] https://www.argos.co.uk/product/7623909?clickPR=plp:6:323
[2] https://amzn.eu/d/bVCBLv3
Its also for free to repair a hole in the hand after trying to knock a hole into the wall ;)
Not free if your time has a value :)
This is true in pretty much every western country sadly.
TV's are really absurdly cheap (and awful) on the low end, we're not talking about your 60" LG OLED with AI TV here, we're talking: a screen with maybe 720p and a viewing angle of: dead centre.
Hiring a handyman is, what, $100/h in most countries, then there's a minimum call-out fee and materials cost- worse "I don't have the part". You're looking at about $300~ easy.
But for $129 you can get this; https://a.co/d/7cdztf8
50" tv for $239 https://www.amazon.com/dp/B0D4P176B8
Huh, i stand corrected, thanks! I think this brings both options to similar prices, so at the very least the spirit of the quote is true. Here’s my math:
- My handyman changes $50/hour, but if you find a new person maybe they charge $75-$100/hour
- materials are cheap, probably like $50 total for mud and drywall, or a repair kit
- with two hours labor, the total should be somewhere from $150-$250.
- if the handyman won’t accept a job less than 4 hours, the range is $250-$450.
> From the article: If you live in the United States today, and you accidentally knock a hole in your wall, it’s probably cheaper to buy a flatscreen TV and stick it in front of the hole, compared to hiring a handyman to fix your drywall.
This also isn't true?
It costs almost nothing to patch drywall. You can also do this yourself. Unless the point they're making is "TVs are so cheap, you can mount a TV inside of the drywall for less money than it would cost to fix," which also isn't true.
If you have the stuff already.
I had to buy the sanding pole, the joint compound, the putty knife, and the paint the other day. A TV would definitely have been cheaper.
More like $20.
In America, all of that other than the paint is available at the dollar tree. You're looking at ~$7 to fix the hole (spackle, mesh tape, trowel, sandpaper, paintbrush) and $12 to buy a pint can of matched paint, as long as the hole is smaller than your fist.
Larger you would need to add in a $20 patch panel of sheetrock, a razor knife or other sheetrock saw (could probably use the bread knife in a pinch) and a hammer and nails, so closer to $50 all in.
> so closer to $50 all in
https://www.amazon.com/toshiba-fire-tv-32-inch-class-v35-ser... is $75 right now. On Prime Day I saw a similar one for $35.
I was curious so I did a quick check.
I can buy a flat screen TV from Walmart for $74. A handyman to come to my house is a minimum of $150.
So the parent comment is true, buying a TV is cheaper than hiring a handyman to fix the drywall.
but you need to put the TV on the wall
VESA mounts and installation are not cheap
it's really an apples and or ciders comparison
that said, now I realized I have a door with two pictures screwed to it, because we punched a hole through it more than a decade ago during a house party, and that specific door is out of production, and it's the door to the storage room so it was (and still is) the perfect solution :)
Walmart has $22 mounts that will work for that $74 TV, and Walmart has TV installation service for $79 (basic mounting of a customer supplied TV with a customer supplied mount).
A TV wall mount can be had for $30, or even little less. They're extremely simple to install. All you need is a basic drill and 20 minutes.
Also, the hole in the wall will make it easier to locate the studs onto which you mount.
Try hiring someone to do the job. You won’t get out for less than a few hundred.
"Some rich guy has been convinced to invest heavily in $NEWTECH, therefore any article voicing skepticism about $NEWTECH must be biased."
This does not follow. Being a wealthy and high-profile investor does not meant that Horowitz understands the technology.
It certainly does not mean that there are no valid criticisms of the technology.
> But that’s not paradoxical at all. That’s basic supply and demand curve from economics 101. It’s exactly what you’d expect.
The key part is the "demand more than eclipsed the cost savings" bit.
The cotton gin is another well known example, labor per unit down, labor for the whole industry up.
> Whereas if you take the extraordinarily difficult step of opening Wikipedia (https://en.wikipedia.org/wiki/Jevons_paradox), you understand the real paradox. Jevons observed that when steam engines get more efficient and individually use less coal, we end up deploying many more steam engines in many new industries and use cases, increasing coal consumption overall.
What difference do you see in this paragraph compared to the one before? Steam engines get more coal-efficient and that means there are more of them burning more coal. This is two observations:
- The cost of steam engines goes down, and the quantity sold goes up.
- The value of coal goes up (it produces more energy per ton), and the quantity sold goes up.
Those results are just as basic as "the cost of coal goes down, and the quantity sold goes up", which is the statement you open by criticizing. I agree there's nothing surprising about it. But there's nothing surprising about the other statements either. The paragraph two statements are both demand increases†, while paragraph 1 is a supply increase, but they all happen on the same X-shaped graph from economics 101. (Or rather, two of them happen on an X-shaped graph measuring supply and demand for coal, and one of them happens on an X-shaped graph measuring supply and demand for engines.)
† "The cost of steam engines goes down" might sound like a supply increase, but the cost that's dropping here is the operating cost and not the manufacturing cost, meaning that, at a given price, more people are willing to buy a steam engine than previously.
> Steam engines get more coal-efficient and that means there are more of them burning more coal.
This misses the key part that actually makes it Jevons Paradox: the reduced cost of operating the machine opened up new industries and applications that massively grew which made overall usage increase.
Something could get more efficient and still have the same number of units in the end. Washing machines have gotten more efficient over time and yet it's not like we have tons more washing machines in the US.
That would require perfectly inelastic demand. It's not a realistic assumption... in any context. Its absence certainly doesn't qualify as a "paradox".
> Washing machines have gotten more efficient over time and yet it's not like we have tons more washing machines in the US.
Do you have any numbers for this?
> Do you have any numbers for this?
Strange you'd really push back on this part. Do you now own multiple washing machines at home because they're more efficient? Seems like a pretty obvious take to me but ok let's go.
Looking at census statistics of durable goods it was in the 80s over 25 years ago.
https://www2.census.gov/programs-surveys/well-being/tables/1...
Nearly 20 years later it was still in the 80s.
https://www.prnewswire.com/news-releases/us-census-bureau-da...
Doesn't seem like there's much room to go up from here unless people are going to start owning multiple for some reason.
Meanwhile they became a lot more energy and water efficient.
> In 2000, 87% of all clothes washers consumed at least 600 kWh/yr. By 2019, 85.4% consumed fewer than 200 kWh/yr.
> The average annual unit energy consumption (UEC) of clothes washers decreased 83.9% between 2000 and 2019
https://oee.nrcan.gc.ca/publications/statistics/aham/2019/wa...
For many things the initial or operating costs aren't what's constraining the market for the product. Clothes washers could be cut in half in cost tomorrow but it's not like I'm now rushing out to go buy twice as many. They could go to only require 5 Whr's to run a load but I'm still going to have about as much laundry. We've pretty much tapped out that market in developed countries, the only really new thing was all in one units becoming more economical to a few percent more of the market that didn't have the plumbing or space availability before.
If we were all too stupid to think of new uses for the steam engine we wouldn't have massively increased their usage by making them cheaper to build and cheaper to operate. We'd have just saved more money and used a bit less coal to do the same things we were doing before. If there isn't a market for the things that steam engine enabled, it doesn't end up becoming Jevons Paradox.
Looking at the Jevons Paradox with spreadsheets and accounting, it wasn't just that the spreadsheet made accounting work so much faster and easier, it's that then the effective output per cost per hour for an accountant dropped massively enabling people to hire part time accountants en masse that increased the demand for accountants. If businesses that never would have bothered for accountants didn't come in and effectively revolutionize what it meant to be an accountant all it would have done was making it cheaper for the same firms to do the same work with fewer people.
as work / token (engine efficiency) increases, because agents are getting smarter, my agent (locomotive) count is increasing, because I find more uses for smarter agents, and my overall token use is increasing.
what's odd is we must take this back to literal coal in order to complete the analogy. tokens are ethereal; energy inputs are not.
cost / token is decreasing while model performance stays roughly the same or gradually improves. cost is a rough proxy for tons of coal consumed.
I think smarter models is what's driving greater coal use overall.
You’re right it’s not a real paradox. It simply states that movements of a/b (efficiency of steam engines input coal / output work, price of coal per kilo of coal etc.) doesnt tell us anything about a and b. Easy maths. But for humans economics usually it means demand increasing so a and b increasing. It’s only a paradox for humans who focus on numbers too much
Oh that's a fine example of the Baader-Meinhof Phenomenon. Read this comment first, went to read one of the other top links on HN (https://news.ycombinator.com/item?id=45804377) and which Paradox shows up? Jevons'!
Was thoroughly confused which article this comment belonged to for a second.
I think that, in a typical HN fashion, this article was submitted after reading the other one, like: here you get another take on this topic.
thing is these a16z guys have money & try to build a narrative around that money.
they did the same in the crypto days - probably most of their narratives were taken down.
then when they finally dumb their investments on the stock market - the market within a year or 2 sees the ruse and stock goes tumbling down. they would've cashed out.
> Jevons observed that when steam engines get more efficient and individually use less coal
And your basic supply and demand curve indicates that, at least until the supply is diminished in kind which isn't going to happen immediately given the all the constraints at play, using less coal leads to coal becoming cheaper and more readily available.
> This seems like a basic oversight
Not really. Improvements to the steam engine being why we got "cheaper and faster at producing coal" is immaterial. The added detail you've given is an interesting aside, I suppose, but doesn't change anything about the original premise. The only oversight is you not realizing this, perhaps?
Honestly I feel like Jevons Paradox is a distinctly unhelpful way of phrasing phenomena, designed to create a feeling of novelty where there's none.
It's completely obvious that if you need energy, and you have energy source A and B (lets say natural gas and oil), people will use the one that is cheaper.
Oil became cheaper because a new supplier entered the market, and people started using more of it! Jevons Paradox!
The other explanation is that people have an outsize demand for a resource, and are actively making efforts to make it cheaper so they can use more of it, then when it gets cheaper, they use more of it.
Transistors got cheaper and we are using more of them! Jevons Paradox!
The paradox isn't that as a good becomes cheaper we're using more of it. The paradox is that as a good becomes cheaper we're spending more on it.