I just can’t handle this guy’s “it’s a tulip bubble!” approach when just yesterday Thinking Machines released a demo which would have been taken as literal magic if it was shown 20 years ago.
The tech is astonishing and Ed spends 98% of his writing acting like the industry is a scam.
This entire article spends a lot of time hiding the ball and conflating what I'd distinguish as "data centers" vs. "data center capacity". Existing data center capacity is being allocated to AI because that's where the demand is, this is in addition to future data center capacity still in the process of being built. Microsoft and others are playing word games and combining both numbers for PR purposes, which I think Ed knows and is being obtuse about.
It's also annoying how he keeps using the term "bubble" as a sleight-of-hand synonym for, I don't know, "useless and fake". Tulip valuations vastly outgrew their actual value and never made it back up. The Dot Com bubble was just early and investment outpaced real-world adoption. Nobody looks back on dot com and thinks claims like "nearly all commerce and social interaction will move online" are silly because that's exactly what happened. AI investment might be at outsized levels right in this moment, it remains to be seen, but even if it doesn't get any better from here it's already insanely useful. That won't "pop" in any meaningful way that he predicts.
EDIT: To say nothing of the overwrought 2023-era comparisons to cryptocurrency. It was obvious at the start that crypto was a solution in search of a problem, shaded by the fact that most advocates benefited strongly from convincing everyone else how valuable of an idea it was. Anybody has been able to spend 20 minutes with free ChatGPT for years now and immediately start to grasp the real-world applications of tech that genuinely replaces a substrate of knowledge work.
As someone who's admittedly anti-AI, what knowledge work does it replace? It seems to me it supplements some knowledge work, while outsourcing actual intelligence to the human operator.
IMO it seems like most AI intelligence is just a Clever Hans situation: the AI produces a stream of responses, and the human selects the one that is correct, then they conclude that the AI is intelligent.
Let me prefix by saying that I'm solidly in some kind of AI middle ground. I think people who are fully outsourcing everything they do to AI are insane, and I think people who have planted their feet and are pretending AI is useless are also insane.
The way I think about it is that a lot of what we considered knowledge work isn't anymore. In "the before times", I would have considered it knowledge work to know how to dig into an unfamiliar code repo or long document and produce a useful summary of the information within, or identify which parts of a codebase are applicable to a given problem I'm trying to solve. AI turns semantic search up to 11; you can point it at an unfamiliar repo and say "what do I have to touch to make this work" and get a 90% accurate result. That's insane magic. I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop, then we're not there yet, but it keeps eating away at more and more of the basic pieces.
> I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop.
That's where I put it personally, because of humans' limited amount of useful focus during a work day.
Anything that requires human attention will take some of that resource, and don't think models' rate of improvement will be fast enough to overcome that in the near future. Reviewing an output that is 99%, 99.9%, or 99.99% correct all take about the same amount of time, so the output needs to be correct enough not to need review before any knowledge work is replaced.
I’m afraid your numbers, all over 99%, are anchoring the conversation to an unreasonably high quality level.
I would have personally gone for 75%, 85% and 95%, which are all still best case scenario answers.
Had I taken on chatbot advice on electronics or chemistry I’d have died every couple of weeks (doing some hands-on real world R&D in my basement as a distraction from software).
"Insanely useful" doesn't mean hundreds of billions of dollars of upfront investment from our largest corporations is prudent especially when proponents of AI don't seem to approach the risk honestly. The argument isn't just that AI is going to take over the world, it's that the consequences of massively over-investing in it could lead to a broader economic collapse more like the Great Depression than the Dot Com bubble. I'm not going to predict what might happen, but I don't think the Dot Com bubble is necessarily predictive of any outcome here, either.
This is not money that would otherwise be going to agriculture or energy; one of the few salient points that Ed makes is that a lot of these deals are circular and based on invented money. I think the comparison to the dot com bubble is sound purely because basically every DC investment that exploded would have been legacy-defining investments if they'd been able to stay solvent until like 2010. Some companies and some people are going to lose, but they're not going to have been wrong, just badly timed.
Tech companies sitting on godlike amounts of money and being able to invest such huge sums in AI at a nascent stage is worrisome. There is absolutely nothing to indicate that this is anything close to an optimal distribution of investment in the economy. It's not a plus that some of these companies are playing an investment shell game. They don't hand out awards for badly timed investments, they hand out bankruptcy rulings. It's fine and good if a few speculative companies go bust. Nobody will think it's fine when companies ingrained into the fabric of the economy do. At the end of the day it's clear we've learned very little from previous crashes and I think it's a false assumption to think we'll suffer equivalently for it.
See data centres being constructed in areas with green energy, forcing countries to revert to petro and update power grids to accommodate FUTURE "needs" and governments discounting power bills due to... yeah, due to what, actually?
Also watch power bills rising in USA due to faulty leadership, and power lines above-ground constantly at risk during the climbing number of destructive hurricanes.
Think bigger and with broader out-reach: All of these companies are forging deals left and right, and - let me guess - some as part of threat-package deals "suggested" by the present US trade administration.
Soon, very soon, many of these US/EU-data centres will be forcefully discontinued as part of growing EU-independence. And good riddence, too.
Yes, I think AI is an industry in need of a prodcuct, and they are minions of the anti-christ of Capitalism and Consumer-Slaverism. This is already a failed project, as it is done without the ethics of collectivity (to the benefit of "more than me"), to paraphrase Kierkegaard. Me, you and All. Only power and enrichment in this end-time, when care could have leveled the plane and touched us all equally GOOD.
People vastly under estimate how hard it is and how much time and money it takes develop anything that doesn't already exist in a dozen places in a given permitting jurisdiction.
You want to build another cookie cutter McMansion, pay through the hoops, break dirt at 6mo and be done in a year.
You want to build a an X over Y apartment building across the street from the other one, pay through the hoops, break dirt at 2yr, be done in 3yr.
You want to build the first X over Y pods and bugs apartment in a given jurisdiction you will be stuck in permitting hell paying through hoop after hoop after hoop, sometimes the same hoop more than once, for 3-5yr right off the bat while the system fights with you.
You can't imagine how terrible anything industrial is. The system is basically only set up to be able to easily approve housing and retail and "easy" still takes years. The amount of money it takes to get something popular approved is insane never mind how much something unpopular like a data center would cost.
Of course every party is playing fast and loose with terminology and capacity. Investors like "we have <big number> and we're adding even more" capacity much more than "we already had <big number> and we've only added <small number> but god willing we'll have <really huge number> in 5yr if we don't go bankrupt first"
I just can’t handle this guy’s “it’s a tulip bubble!” approach when just yesterday Thinking Machines released a demo which would have been taken as literal magic if it was shown 20 years ago.
The tech is astonishing and Ed spends 98% of his writing acting like the industry is a scam.
This entire article spends a lot of time hiding the ball and conflating what I'd distinguish as "data centers" vs. "data center capacity". Existing data center capacity is being allocated to AI because that's where the demand is, this is in addition to future data center capacity still in the process of being built. Microsoft and others are playing word games and combining both numbers for PR purposes, which I think Ed knows and is being obtuse about.
It's also annoying how he keeps using the term "bubble" as a sleight-of-hand synonym for, I don't know, "useless and fake". Tulip valuations vastly outgrew their actual value and never made it back up. The Dot Com bubble was just early and investment outpaced real-world adoption. Nobody looks back on dot com and thinks claims like "nearly all commerce and social interaction will move online" are silly because that's exactly what happened. AI investment might be at outsized levels right in this moment, it remains to be seen, but even if it doesn't get any better from here it's already insanely useful. That won't "pop" in any meaningful way that he predicts.
EDIT: To say nothing of the overwrought 2023-era comparisons to cryptocurrency. It was obvious at the start that crypto was a solution in search of a problem, shaded by the fact that most advocates benefited strongly from convincing everyone else how valuable of an idea it was. Anybody has been able to spend 20 minutes with free ChatGPT for years now and immediately start to grasp the real-world applications of tech that genuinely replaces a substrate of knowledge work.
As someone who's admittedly anti-AI, what knowledge work does it replace? It seems to me it supplements some knowledge work, while outsourcing actual intelligence to the human operator.
IMO it seems like most AI intelligence is just a Clever Hans situation: the AI produces a stream of responses, and the human selects the one that is correct, then they conclude that the AI is intelligent.
Let me prefix by saying that I'm solidly in some kind of AI middle ground. I think people who are fully outsourcing everything they do to AI are insane, and I think people who have planted their feet and are pretending AI is useless are also insane.
The way I think about it is that a lot of what we considered knowledge work isn't anymore. In "the before times", I would have considered it knowledge work to know how to dig into an unfamiliar code repo or long document and produce a useful summary of the information within, or identify which parts of a codebase are applicable to a given problem I'm trying to solve. AI turns semantic search up to 11; you can point it at an unfamiliar repo and say "what do I have to touch to make this work" and get a 90% accurate result. That's insane magic. I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop, then we're not there yet, but it keeps eating away at more and more of the basic pieces.
> I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop.
That's where I put it personally, because of humans' limited amount of useful focus during a work day.
Anything that requires human attention will take some of that resource, and don't think models' rate of improvement will be fast enough to overcome that in the near future. Reviewing an output that is 99%, 99.9%, or 99.99% correct all take about the same amount of time, so the output needs to be correct enough not to need review before any knowledge work is replaced.
I’m afraid your numbers, all over 99%, are anchoring the conversation to an unreasonably high quality level.
I would have personally gone for 75%, 85% and 95%, which are all still best case scenario answers.
Had I taken on chatbot advice on electronics or chemistry I’d have died every couple of weeks (doing some hands-on real world R&D in my basement as a distraction from software).
"Insanely useful" doesn't mean hundreds of billions of dollars of upfront investment from our largest corporations is prudent especially when proponents of AI don't seem to approach the risk honestly. The argument isn't just that AI is going to take over the world, it's that the consequences of massively over-investing in it could lead to a broader economic collapse more like the Great Depression than the Dot Com bubble. I'm not going to predict what might happen, but I don't think the Dot Com bubble is necessarily predictive of any outcome here, either.
This is not money that would otherwise be going to agriculture or energy; one of the few salient points that Ed makes is that a lot of these deals are circular and based on invented money. I think the comparison to the dot com bubble is sound purely because basically every DC investment that exploded would have been legacy-defining investments if they'd been able to stay solvent until like 2010. Some companies and some people are going to lose, but they're not going to have been wrong, just badly timed.
Tech companies sitting on godlike amounts of money and being able to invest such huge sums in AI at a nascent stage is worrisome. There is absolutely nothing to indicate that this is anything close to an optimal distribution of investment in the economy. It's not a plus that some of these companies are playing an investment shell game. They don't hand out awards for badly timed investments, they hand out bankruptcy rulings. It's fine and good if a few speculative companies go bust. Nobody will think it's fine when companies ingrained into the fabric of the economy do. At the end of the day it's clear we've learned very little from previous crashes and I think it's a false assumption to think we'll suffer equivalently for it.
[dead]
Look outside of USA.
See data centres being constructed in areas with green energy, forcing countries to revert to petro and update power grids to accommodate FUTURE "needs" and governments discounting power bills due to... yeah, due to what, actually?
Also watch power bills rising in USA due to faulty leadership, and power lines above-ground constantly at risk during the climbing number of destructive hurricanes.
Think bigger and with broader out-reach: All of these companies are forging deals left and right, and - let me guess - some as part of threat-package deals "suggested" by the present US trade administration.
Soon, very soon, many of these US/EU-data centres will be forcefully discontinued as part of growing EU-independence. And good riddence, too.
Yes, I think AI is an industry in need of a prodcuct, and they are minions of the anti-christ of Capitalism and Consumer-Slaverism. This is already a failed project, as it is done without the ethics of collectivity (to the benefit of "more than me"), to paraphrase Kierkegaard. Me, you and All. Only power and enrichment in this end-time, when care could have leveled the plane and touched us all equally GOOD.
This is funny to say because EU QOL stays afloat due to cheap imports from China.
People vastly under estimate how hard it is and how much time and money it takes develop anything that doesn't already exist in a dozen places in a given permitting jurisdiction.
You want to build another cookie cutter McMansion, pay through the hoops, break dirt at 6mo and be done in a year.
You want to build a an X over Y apartment building across the street from the other one, pay through the hoops, break dirt at 2yr, be done in 3yr.
You want to build the first X over Y pods and bugs apartment in a given jurisdiction you will be stuck in permitting hell paying through hoop after hoop after hoop, sometimes the same hoop more than once, for 3-5yr right off the bat while the system fights with you.
You can't imagine how terrible anything industrial is. The system is basically only set up to be able to easily approve housing and retail and "easy" still takes years. The amount of money it takes to get something popular approved is insane never mind how much something unpopular like a data center would cost.
Of course every party is playing fast and loose with terminology and capacity. Investors like "we have <big number> and we're adding even more" capacity much more than "we already had <big number> and we've only added <small number> but god willing we'll have <really huge number> in 5yr if we don't go bankrupt first"
Ah, it’s the Hysterical Guy Who is Wrong About AI again.