So Bing was using GPT-4 after all. That explains why it felt noticeably more capable than chatGPT. Still, this advance seems like a less revolutionary leap over GPT-3 than GPT-3 was over GPT-2, if Bing’s early performance is a decent indicator.
Lost Futures
As Adam said, trending with Moore’s Law is far slower than the previous trajectory of model scaling. In 2020 after the release of GPT-3, there was widespread speculation that by the next year trillion parameter models would begin to emerge.
Georgists, mandatory parking minimum haters, and housing reform enthusiasts welcome!
Recently I’ve run across a fascinating economics paper, Housing Constraints and Spatial Misallocation. The paper’s thesis contends that restrictive housing regulations depressed American economic growth by an eye-watering 36% between 1964 and 2009.
That’s a shockingly high figure but I found the arguments rather compelling. The paper itself now boasts over 500 citations. I’ve searched for rebuttals but only stumbled across a post by Bryan Caplan identifying a math error within the paper that led to an understatement(!) of the true economic toll.
This paper should be of great interest to anyone curious about housing regulation and zoning reform, Georgism, perhaps even The Great Stagnation of total factor productivity since the 70s. (Or just anyone who likes the idea of making thousands of extra dollars annually.)
If there’s interest, I’d like to write a full-length post diving deeper into this paper and examining its wider implications.
I’m also quite sympathetic to the idea that another AI winter is plausible, mostly based off compute and data limits. One trivial but frequently overlooked data point is that GPT-4 was released nearly three years after GPT-3. In contrast, GPT-3 was released around a year after GPT-2 which in turn was released less than a year after GPT-1. Despite hype around AI being larger than ever, there already has been a progress slowdown relative to 2017-2020.
That said, a big unknown is to what extent specialized hardware dedicated to AI can outperform Moore’s Law. Jensen Huang sure thinks it can:
So obviously, computing has advanced tremendously and the way that’s happened, of course, is a complete reinvention of how computers write software, the computer architecture of it, and the computer runs software. Every single layer from the chip to the system to the interconnect to the algorithms, all completely redesigned and so this way of doing full-stack computing as you projected out ten years, there’s no question in my mind, large language models and these very large language models will have an opportunity to improve by another factor of a million. It just it has to be full stack.
That said, the economy is absorbing AI much slower than it is progressing and even if frontier progress halts tomorrow, investment may still be buoyed by the diffusion of the current models. It’s hard to argue that current models aren’t powerful enough to have economic value and won’t get less expensive as time progresses, regardless of how the frontier moves.
Found an obscure quote by Christiaan Huygens predicting the industrial revolution a century before its inception and predicting the airplane over two hundred years before its invention:
The violent action of the powder is by this discovery restricted to a movement which limits itself as does that of a great weight. And not only can it serve all purposes to which weight is applied, but also in most cases where man or animal power is needed, such as that it could be applied to raise great stones for building, to erect obelisks, to raise water for fountains or to work mills to grind grain …. It can also be used as a very powerful projector of such a nature that it would be possible by this means to construct weapons which would discharge cannon balls, great arrows, and bomb shells …. And, unlike the artillery of today these engines would be easy to transport, because in this discovery lightness is combined with power.
This last characteristic is very important, and by this means permits the discovery of new kinds of vehicles on land and water.
And although it may sound contradictory, it seems not impossible to devise some vehicle to move through the air ….
While ultimately land, water, and air vehicles wouldn’t be powered by Huygens’s gunpowder engine, it remains a remarkably prescient forecast. It should also give AI researchers and other futurists some hope in their ability to predict the next technological revolution.
Is it true that 19th-century wheelwrights were extremely highly paid?
I’m quite skeptical of the claim that wheelwrights made $90 a week in 1880s.
San Francisco Call, Volume 67, Number 177, 26 May 1890: A job listing offers $3.50 a day for wheelwrights. Another offers $75(!) but I suspect this is for a project rather than a daily (or weekly) wage.
San Francisco Call, Volume 70, Number 36, 6 July 1891: Two job listings offer $3 a day for wheelwrights. Another offers $30 to $35 for a “wheelwright: orchardist” but again I suspect this is commission work rather than a daily wage.
San Francisco Call, Volume 96, Number 135, 13 October 1904: Two wheelwright job listings offer a daily wage of $3 and $3.50 respectively. And from the San Francisco Call, Volume 96, Number 46, 16 July 1904, three additional job listings for wheelwrights all offer compensation between $3 and $3.50 a day.
Whew, I think we’ve figured it out.
Even working daily with no rest, an average wheelwright in San Fransisco from 1890-1905 probably made no more than $25 weekly. Of course with a rising wave of mechanization, it’s possible wheelwright wages were previously higher. After all, from 1890 to 1904, their wages do seem to be declining accounting for inflation. And maybe SF wheelwrights were simply paid less than average. Even still, $90 in 1880 seems unlikely for an average American wheelwright.
TL;DR
Wheelwrights in the 1880s almost certainly made far less than $90 a week. Probably not even $45 a week.
17th century Netherlands contains another interesting case. The depletion of peat, a primary energy source for the Dutch between the 16th and 17th centuries, directly contributed to the end of the Dutch Golden Age and economic stagnation, even decline. The Dutch economy could perhaps have continued growing had it embraced coal as peat supplies depleted, but no such switch occurred. According to The Rise and Decline of Dutch Technological Leadership by Karel Davids:
The Dutch succeeded in raising output per capita to an unheard-of extent for a prolonged period of time by making increased use of a stock of energy resources, instead of a flow, in the form of large deposits of peat. Eventually, however, the Netherlands did not escape the ‘limitations experienced by all organic economies’, namely relatively low maximum levels of energy input and productivity growth, given the ‘extreme inefficiency of the process of photosynthesis in converting solar energy into a form accessible to living creatures’. Increased reliance on peat postponed the day of reckoning, Wrigley argues, but it also implied that Dutch industries, thriving for a long time on cheap heat energy, found it difficult to compete once the depletion of peat stocks led to rising prices of fuel. In contrast with eighteenth-century England, the Dutch Republic did not to make a transition to a ‘mineral-based energy economy’, which allowed a outlet from the traditional constraints on energy input and productivity growth.
Davis Kedrosky argues that the Dutch government, rather than market forces, prevented a switch to coal. Nonetheless, this does appear to be an example of major economic damage caused by resource depletion.
I largely agree with the sentiment of your post. However, one nitpick:
The world’s largest protest-riot ever, when measured by estimated damage to property.
This claim is questionable. The consensus is that the economic cost of the George Floyd Protests was between one and several billion. Perhaps it was the most expensive riot in US history (though when inflation-adjusted the LA riots may give it a run for its money) and the most expensive to be cleanly accounted for economically, but intuitively I would imagine many of the most violent riots in history, such as the partition riots in India and Pakistan, caused more economic damage.
Thanks for the explanation Gwern. Goodhart’s law strikes again!
Altman said there are also physical limits to how many data centers the company can build and how quickly it can build them.
This seems to insinuate a cool down in scaling compute and Sam previously acknowledged that the data bottleneck was a real roadblock.
- 22 Jun 2023 23:05 UTC; 1 point) 's comment on AI #17: The Litany by (
Pretty sure that’s just an inside joke about Lex being a robot that stems from his somewhat stiff personality and unwillingness to take a strong stance on most topics.
Thanks for the detailed and informative response Breakfast! I think I largely agree with your post.
I find it likely that that the coincidence of the Montgolfier brothers’ and Lenormands’ demonstrations in France in 1873 was no accident. There was something about that place and that time that motivated them. If I had to guess, it was something cultural: the idea of testing things in the real world, familiarity with hundreds of years of parachute designs, a critical mass of competitive and supportive energy in the nascent aeronautics space, increasing cultural familiarity with connecting physical intuitions with practical engineering to design”magical” machines.
(1783* you mean.) A revolution in thought definitely aided the invention of the hot air balloon. Novel philosophical ideas and the scientific revolution inspired a more discerning examination of the invention space. But let me ask you this, do you believe the hot air balloon could not have been invented prior to these cultural ideas and parachute design knowledge? My intuition says no, especially given that the Montgolfiers’ first balloon prototype was just a large sky lantern made of thin wood and taffeta lifted by burning paper.
IMO, the hot air balloon is an invention that had a fair probability of being invented anytime after the invention of the sky lantern but simply failed to materialize until the scientific revolution and aeronautics pushed said probability near 100%.
I’m skeptical. Guzey seems to be conflating two separate points in the section you’ve linked:
TFP is not a reliable indicator for measuring growth from the utilization of technological advancement
Bloom et al’s “Are Ideas Getting Harder to Find?” is wrong to use TFP as a measure of research output
The second point is probably true, but not the question we’re seeking to answer. Research output does not automatically translate to growth from technological advancement.
For example, the US TFP did not grow in the decade between 1973 and 1982. In fact, it declined by about 2%. If – as Bloom et al claim – TFP tracks the level of innovation in the economy, we are forced to conclude that the US economy regressed technologically between 1973 and 1982.
Of course such conclusion is absurd.
Is it absurd? I’m not so sure. Between ’73 and ’82 the oil shock led to skyrocketing energy prices. Guzey acknowledges this economic crisis but goes on to claim that the indicator must be bad since semiconductors got better, crop yields improved, and life expectancy improved. And he’s right, for Bloom’s paper, this is a major discrepancy. TFP is not a good measure of research output. However, TFP roughly measures an economy’s technological capacity given current restraints.
America in ’73 was more productive than America in ‘82 because a key technological input (energy) was significantly cheaper in ’73 than it would be for most of the following decade while the technological advancements made during the same period were not enough to offset the balance.
Let’s look at the other examples provided:
According to the data provided, France’s TFP peaked prior to the Great Recession and has largely stagnated since. This doesn’t seem surprising given France’s sluggish economic growth since then. French GDP peaked in 2008. Its labor productivity has also barely grown. If one examines the data without holding the bias that tech advancements since 2001 MUST have vastly improved productivity, the results are hardly surprising.
This is harder to explain. According to the data, Italy’s TFP effectively peaked in 1979, remained near this peak until just before the Great Recession, and declined since. Italy’s GDP peaked around the time of the Great Recession and declined since. Nonetheless, its TFP being higher in 1970 than 2019 is shocking. CEPR argues that Italian manufacturing misallocates resources on a massive scale but I’d hesitate to give any firm opinion. Rising energy costs may also play a role? This is worthy of further research, but as Guzey points out, Italy is not on the technological frontier and is a bit of a basket case.
Japan’s TFP in 1990 was higher (a) than in 2009.
Unsurprising. 2009 was an unusually weak year for TFP in Japan given the Great Recession’s effects. Moreover, since the 1990s, Japan has been in its lost decades. Japan’s TFP growth looks more healthy and similar to America’s compared to France and Italy.
(See Italy)
Skipping Sweden and Switzerland as they are small countries.
The United Kingdom’s TFP peaked in 2007, one year before its GDP peaked. Like France, Italy, and Spain, it has yet to recover from the Great Recession.
TFP is NOT a measure of the pure technological frontier. It cannot tell you how much cutting-edge lab research has progressed over time. What it can tell you is how much technological advancement has soaked into the economy. Recessions, market shocks, structural barriers, and other forms of inertia can slow or even regress TFP.
Typically those who subscribe to the premise of The Great Stagnation have a dim perception of progress in the world of bits. The argument goes that digital progress is overhyped in the present day because it’s one of few, if not the only industry still exhibiting fast progress.
Zooming out to the wider economy, GDP growth rates have slowed enormously in the developed world since the early 70s and total factor productivity growth has declined to rates possibly not seen since the dawn of the industrial revolution. If these measurements are indeed still a reasonable metric of technological progress in the digital era then advancements in computing have not managed to match midcentury levels of progress.
If you accept this analysis but still bank on AI as the solution moving forward, that means accepting the notion that computing progress hasn’t lived up to expectations in the past but will lead to tremendous advancements just on the horizon. Given how many times AI has been hyped only to fall into a stagnant winter soon after, I’m not so sure about this.
Yep, just as developing countries don’t bother with landlines, so to will companies, as they overcome inertia and embrace AI, choose to skip older outdated models and jump to the frontier, wherever that may lie. No company embracing LLMs in 2024 is gonna start by trying to first integrate GPT2, then 3, then 4 in an orderly and gradual manner.
Erik Engheim and Terje Tvedt introduced me another important development in Europe that seems connected to the industrial revolution: The Machine Revolution. While Medieval China invented plenty of industrial machines, including the first water-powered textile spinning wheel, by the high middle ages Western Europe was using more water and wind power per capita than anywhere else in history.
...watermills in 1086 did the work of almost 400,000 people at a time when England had no more than 1.25 million inhabitants. That means doing as much work as almost 30 percent of the English population.
Cultural explanations are often given for this divergence but in recent years, geography has been given greater focus. Western Europe had waterways more naturally suited for the use of water-power than the rest of Eurasia.
While water and wind power had been harnessed for centuries prior, this intensification of capital provided Europe with far more output than what would otherwise have been possible and more opportunities for engineers and mechanics to experiment, tinker with, and improve machinery.
It may be that the earliest seeds of industrial revolution were planted as far back as the middle ages.
Interesting, how good can they get? Any ELO estimates?
Thanks for the response jmh!
One idea might be that it should have been invented then IF the idea that air (gases) were basically just like water (fluids).
I dunno if this is an intuitive jump but it seems unnecessary. Sky lanterns were built without knowledge of the air acting as a fluid. I don’t see why the same couldn’t be true for the hot air balloon.
But there would also have to be some expected net gain from the effort to make doing the work worthwhile. Is there any reason to think the expect value gained from the invention and availability of the balloon was seen as anything more than a trivial novelty or toy (such as the Chinese seemed to think)?
As I understand it, expectations for the hot air balloon were placed too high rather than too low. In the 1600s, Francesco Lana de Terzi envisioned that a hypothetical airship (which he deemed impossible) could break sieges (ofc airships are not the same as hot air balloons, but at the time there was no distinction). A very valuable use case. After the invention of the hot air balloon, lofty expectations continued for some time. From Wikipedia, “The military applications of balloons were recognized early, with Joseph Montgolfier jokingly suggesting in 1782 that the French could fly an entire army suspended underneath hundreds of paper bags into Gibraltar to seize it from the British. Military leaders and political leaders soon began to see a more practical potential for balloons to be used in warfare; specifically in the role of reconnaissance.”
After all, a balloon is not much like a ship which can be steered and the value of higher ground limited to just how far one can see clearly, and with sufficient detail.
This wasn’t known prior to the invention of the hot air balloon. Bartolomeu de Gusmão, who allegedly built a prototype of something similar to a hot air balloon in the early 1700s expected it to be steerable like a ship.
The Archimedes example might be an easy case, but I’m wondering if there are not things to look into regarding the motivations for the work on an invention at the time that offer some type of change in the “environment” (social or intellectual/level of knowledge) that point to why no one did something we now think of as obvious.
The scientific and budding industrial revolution motivated a “spirit of invention”. The idea of being an inventor by profession took root and led to more people taking a detailed look at the invention space. IMO, this shift in thinking turned the hot air balloon from an invention that some lone inventor with sufficient capital could have invented into a statistical inevitability.
Guzey goes on to give other takes I find puzzling like the following:
If Google makes $5/month from you viewing ads bundled with Google Search but provides you with even just $500/month of value by giving you access to literally all of the information ever published on the internet, then economic statistics only capture 1% of the value Google Search provides.
He already has his conclusion and dismisses arguments that reject it. “Of course the internet has provided massive economic value, any metric which fails to observe this must be wrong.” What is the evidence that Google Search provides consumers with $500/month of value? The midcentury appliances revolution alone saved families 20 hours or more of weekly labor. No one argues that the digital revolution hasn’t improved technological productivity, economists cite it as the cause of the brief TFP growth efflorescence from the mid-90s to the early 2000s. But Guzey seems to think its impact is far larger and imagines scenarios to support this claim.
This is my first post on LessWrong as well as my Substack. Been sitting on this post for a while but finally dug up the courage to publish it today. Any feedback would be greatly appreciated!