It’s not a totally unreasonable assertion, it’s the implication of the assertion that we are uncomfortable with. There is no reason for the models to stop their improvements in the near future.
Sure there is.
1. The cost of each new generation of training runs appears to be rapidly rising
2. The Trump admin just told the leading model to stop making it available to non-Americans, which in practice meant stop providing it at all
3. The factories to make the hardware are hitting bottlenecks, and while they've currently been navigated around, there's never a guarantee the next one will be
Currently I'm wondering at what point the direct impact on the US energy supply gives the US a taste of Baumol's cost disease as AI companies continue to outbid everyone else for electricity.
Implausible while Trump remains in office. He hates renewables, shuts them down even when doing so actively costs money.
Between AI hallucinated content and the politicisation of the numbers, I'm not sure how much AI compute capacity is being planned right now; would you accept a claim of 300 GW? It's a number I heard recently.
Given the capacity factor of PV, even China would have to think carefully before supplying that much PV over the next few years (300 GW avg ~= 3TW nameplate).
(Not sure about wind, wind's CF seems to vary between years).
You speak as if "improvements to models" is just function of time, and resources are infinite.
Models keep improving as long as there are resources to allow for larger and larger datacenters, if we hit a scientific breakthrough once LLM technology become the bottleneck, if the economy is infinite to allow infinite growth, and (geo)politics is not a thing to worry about. Or we discover ASI, machine improve themselves and we reach the technological singularity.
I know everybody is drinking the kool aid by the gallon, but can we maintain a little bit of objectivity?
Granted, we don't know when the S-curve will inflect, but predicting too great an outcome is just as silly as discounting it altogether.
How does argrar industry and tech industry compare as share of gdp in the US?
Food is the 3rd largest expenditure in most households, after housing and transportation.
It just shows how much the automation has impacted agriculture and the food industry. Sure, there're rural farms that apply 200 yo technology. But e.g. the grain production and farming are incredibly efficient at scale. So, it's not that costly for as a humanity to feed 8 billion people (at a varying level, of course).
Remove 20% of food supply, and watch prices explode, global unrest, and famine take place.
In the places where famine remains a problem, it's due to political issues, not that we can't grow enough.
And growing all that food requires a tiny workforce compared to 400 years ago before the Agricultural Revolution. AI might extend such a massive reduction in labour requirements to many other industries.
Mmmmmh
> We can grow far more food than we need and we stop doing so simply because the low prices mean it's not economically viable.
So, it's not a solved problem. Last time I checked we have plenty of people in several parts of the world with difficulties to access the required level of food to be healthy.
Half. This depends on there being a reliable source of cheap fertiliser, which would be much more secure if not for the situations regarding Hormuz and Russia.
You need lots of money to be able to buy the tech you need to do so. And you can't exactly earn that from not using the tech, since foreign (or even local) competition will slaughter you on prices. And if you do make it, you're stuck with a low-margin race to the bottom on price.
The political issues are still there so I really don't think we can call that a solved problem.
I spend 0.2% on AI. Exactly one subscription.
It's not unthinkable that trend continues (even if it's rationalizing at the moment), and moves over into other fields as well.
my point is that the amount of calories a person needs is limited, and the efficiency is non-decreasing, so the per capita spending has an upper bound
"ai" does not have such an upper bound
Reasoning and RAG is amazing already and is a productivity gain but I'm yet to be convinced GenAI is anything but a slop machine.
#startflamingmenow
AI is useful as a search & information synthesis tool, and as a dev tool.
The problem is, when has a dev tool ever command such ridiculous valuations and investment in infrastructure?
The market is going to realize that yes, it's useful, but no, it's not over $1T useful.
I think the issue is the reality that most life is worth a lot less (in US Freedom units) than some software running doing absolutely nothing truly valuable for anyone.
This is really starting to feel like the pets.com era again.
There are no growth opportunities in any other industry (except healthcare due to disastrous demographics), where else are people going to invest?
The problem is an absolute lack of vision on the part of those holding the capital.
Meeting the challenges of climate change could hold huge opportunities. Look at China’s massive expansion in renewables, look at the expansion of renewables in the US despite political headwinds.
Have some imagination, break out of your echo chamber. AI ain’t the only game to be played.
Vision is seeing a change that could be made. “I could be richer” is about as banal as a vision could get.
The money chasing investments is orders of magnitude larger than the money people have on their pockets to spend. As a consequence, the only profitable thing to do is sell capital goods to make business and there is no profit on selling actually useful things.
China is in a different reality in large part because of their capital barriers that stop money from flowing in. Countries with bad reputation are also less affected.
What the GP gets wrong is that none of this makes AI a good business. Instead, it makes Nvidia a good business, but that's not news.
I don't think that will be Musk. He'll probably pull out significant resource from all this financial engineering relatively quickly. Probably via more financial engineering.
- We get an open source Opus 4.8 equivalent and pair it with an open source coding agent
- Running this OS stack becomes cheaper than what frontier model providers charge (see OS model prices on OpenRouter vs. frontier lab prices)
- This happens across verticals (i.e. not just software)
The first “DeepSeek moment” didn’t do much damage back in the days, but I wouldn’t be surprised if a similar moment becomes a lasting, effective, cheaper alternative.
DeepSeek v4 Flash/Pro also exist, they are open weight and on par with Sonnet, just a bit below Opus. Again: practically useful and sufficient for 80% of things most people actually do. And most of the remaining 20% are benchmarks designed to push the limits, not productive work.
Using these already is way cheaper than your typical Claude API prices. What's still missing is a) mindshare - everyone still thinks "claude = coding" and everyone thinks he/she really needs the very best models because he/she is doing such incredibly complex stuff - and b) someone pushing such a stack as a convenient solution for corporations to easily dump their token money into, complete with user management, enrollment, monitoring, all that enterprisey stuff you need if you want to sell to, well, enterprise customers.
That is an interesting point. If there are higher concerns, copyright law is easily ignored, and only one person needs to get access to the data once.
The thing with dotComs is that they didn't have THIS level of unsustainable financing burn, and a tangible issue of token processing cost that has no magic wand coming with the current practical limits of Moore's law.
Look at electricity, the world of 1900 could not create enough electricity or even conceive of how to add enough to meet 1950s needs. But we made it incredibly cheaper to produce, but also created a lot more, and boy do we have so much more use of electricity now. And it's not that expensive for a human to pay for their needs (not free, its not cheap for poor people but it's still gotten cheaper).
Why limit yourself to one planet? Space is infinite ;-)
AI might eventually provide $26T worth of value, but if it captures anywhere close to that amount of revenue that'll indicate a failure of the free market economy. Competition and open source will have failed and the oligarchy has won.
(Either that, or inflation will have made $26T a relatively smaller number).