Marc Andreessen and I talk AI at an a16z American Dynamism event
a16z has issued the talks from that event, and we are issuing it too, as a bonus episode of CWT. But note it is shorter than usual, and not the typical CWT format — this was done for an audience of actual DC human beings!
Excerpt:
COWEN: Why is open-source AI in particular important for national security?
ANDREESSEN: For a whole bunch of reasons. One is, it is really hard to do security without open source. There are actually two schools of thought on information security, computer security broadly, that have played out over the last 50 years. There was one school of security that says you want to basically hide the source code, and you want to hide the source code precisely. This seems intuitive because, presumably, you want to hide the source code so that bad guys can’t find the flaws in it, right? Presumably, that would be the safe way to do things.
Then over the course of the last 30 or 40 years, basically, what’s evolved is the realization in the field (and I think very broadly) that actually, that’s a mistake. In the software field, we call that “security through obscurity,” right? We hide the code. People can’t exploit it. The problem, of course, is: okay, but that means the flaws are still in there, right?
If anybody actually gets to the code, they just basically have a complete index of all the problems. There’s a whole bunch of ways for people to get the code. They hack in. It’s actually very easy to steal software code from a company. You hire the janitorial staff to stick a USB stick into a machine at 3:00 in the morning. Software companies are very easily penetrated. It turned out, security through obscurity was a very bad way to do it. The much more secure way to do it is actually open source.
Basically, put the code in public and then basically build the code in such a way that when it runs, it doesn’t matter whether somebody has access to the code. It’s still fully secure, and then you just have a lot more eyes on the code to discover the problems. In general, open source has turned out to be much more secure. I would start there. If we want secure systems, I think this is what we have to do.
Marc is always in top form.
Those new service sector jobs, squatter removal edition
Via the excellent Samir Varma, hail to Handyman Flash Shelton:
After local law enforcement couldn’t help, Shelton spent days dissecting laws around squatters’ rights. He managed to get rid of the women within a day by drafting a lease agreement with his mother designating him the legal resident of the home, then took over the house when the women stepped out one day and barred them from re-entering.
Now he uses his experience to provide squatter removal services for others and has successfully helped several landlords in California reclaim their homes.
“I think it’s just something that is coming to light … and I believe that it’s going to get worse,” Shelton told Fox News. “Squatters’ rights were never intended to allow the takeover of residential maintained properties. So until we make it criminal, it’s just going to keep happening, and people are going to be afraid to rent out or buy.”
In October, a 4,000 square foot, five-bedroom Atlanta home was taken over by squatters who ran an illegal strip club inside on weekends and kept horses on the property, neighbors told WSB-TV. Ultimately, the FBI arrested four people residing in the trashed house.
Another Atlanta resident discovered squatters had broken into her property that she was selling. She said there was prostitution, drug use and $30,000 worth of damage done to her home.
Here is the full story, here is (gated) LA Times coverage.
Steve Levitt is retiring from academia
In latest @CapAndFreedom ep, Steve Levitt talks his career & retiring as @UChicago prof while still leading RISC & Freakonomics. We talk applied economics, Chicago econ dept (price theory failure, macro success, personalities), Freakonomics origin & more https://t.co/bdBKovL7PF
— Jon Hartley (@Jon_Hartley_) March 13, 2024
Wednesday assorted links
1. The largest worms on earth.
2. AI safety is not a model property.
3. Dan Schulz podcast with Nabeel Qureshi, with transcript.
4. African influencers who make it big in Brazil.
7. Be careful what you wish for: “The proposed legislation may force app stores to remove TikTok. But restricting access through web browsers or already-installed apps—which would be necessary to really limit the platform’s reach—would represent another level of intrusive regulation.” (WSJ)
Sentences of note
COVID-19 vaccination reduced the risk of post-COVID-19 cardiac and thromboembolic outcomes. These effects were more pronounced for acute COVID-19 outcomes, consistent with known reductions in disease severity following breakthrough versus unvaccinated SARS-CoV-2 infection.
Here is the full article by Núria Mercadé-Besora, et.al.
The Canadian economy is a worthwhile Canadian initiative
Contrary to some recent claims, the Canadian economy is not falling apart, and now I have a Bloomberg column on that point. Excerpt:
A chorus of doomsayers is pointing out that by some measures, Canadian per-capita GDP is in decline.
While there is genuine room for concern, the bad news is much overstated. Canada has not altered its fundamental and longstanding position: It is somewhat poorer than the US, but it is progressing at a more or less typical pace.
First, a note on the numbers: Canadian GDP per capita has not regressed to where it was in 2014. That metric deflates incomes by producer prices, whereas for actual living standards economists generally agree that a consumer-prices deflator of some kind is more appropriate. Using that metric, Canadian per-capita GDP mostly has been rising since 2014, the pandemic aside, and currently stands modestly above pre-pandemic levels, albeit with a very recent downturn. Yes, Canadian performance could be better, but there’s no reason to be pressing the panic button.
It is also noteworthy that the Canadian economy, historically, does better when measured by median income trends, which have been steadily positive for almost two decades, including since 2014.
As for concerns:
It is a fair to wonder why the Canadian economy, in relative terms, seems to be slipping behind the US. As a general observation, this is true of most developed economies, and perhaps it says more about American virtues than Canadian defects.
More specifically, though, the nature of the tech economy may suggest this widening gap is inevitable, at least for a while. Canadian contributors to the technology scene are numerous, most of all in artificial intelligence. But with some exceptions, these developments have had the most impact in the US. Canada can certainly be proud of the contributions of the brilliant Canadian-Israeli computer scientist Ilya Sutskever, a co-founder of OpenAI, but of course that means he was not working in Canada.
Think of it as the economic equivalent of the US’s exorbitant privilege in monetary policy: The more other economies become technology-centered, the more likely they will help the major tech clusters in the US, which will continue to attract global talent. So it is possible that the US will increase its relative lead over Canada, and many other economies, no matter what Canada does. Perhaps some Canadians will feel bad about that growing gap, but to the extent their country benefits from those same tech innovations, it is good for most Canadians.
I am looking forward to my next Canada trip, when will that be? I don’t think I will make it to the Candidates’ tourney, but I am picking Caruana to win.
Claims about compute
Dell let slip during their earnings call that the Nvidia B100 Blackwell will have a 1000W draw, that's a 40% increase over the H100. The current compute bottleneck will start to disappear by the end of this year and be gone by the end of 2025. After that, it's all about power.
— Andrew Curran (@AndrewCurran_) March 11, 2024
Hard for me to judge this one, but do not underestimate elasticity of supply. And by the way, initial reports on Devin are very positive.
Tuesday assorted links
1. Over 2015-2021, the number of Chinese workers in Africa fell by 64 percent (note the link has too many pop-ups, click only if you have to).
2. Seasonal pollen boosts traffic fatalities.
3. Golden Mall reopens in Flushing, Queens (NYT).
4. Katja Grace and AI worries (New Yorker). And a general update on the AI worries.
5. Henry Oliver on James Joyce.
6. A post-mortem on neoreaction.
7. The Alliance for the Future Manifesto, on AI, by Brian Chau.
Claude 3 Opus Also Fails Steve Landsburg’s Economics Exam
Almost one year ago, Steve Landsburg tried GPT-4 on one of his exams. It failed, badly. I tried out some of the same questions on Claude 3 Opus, by many accounts now the leading AI. It failed, badly. Steve’s exams are very clever. They aren’t technically difficult but they are tricky in the sense that the answer you would get by reading the textbook is almost certainly wrong. Not because the textbook is wrong but because you have failed to really understand. Here’s two examples:
Q. Apples are provided by a competitive industry. Pears are provided by a monopolist. Coincidentally, they sell at the same price. You are hungry, and would be equally happy with an apple or a pear. If you care about conserving societal resources, which should you buy?
A. If you care about conserving societal resources, you should buy the apple provided by the competitive industry rather than the pear provided by the monopolist. Here’s why:
In a competitive industry, firms are price takers and must operate efficiently to survive. They produce at the point where marginal cost equals price. This means they are producing at the socially optimal level, using society’s resources efficiently.
In contrast, a monopolist has market power and can set prices above marginal cost. They produce less than the socially optimal quantity in order to drive up prices and maximize profits. This results in a deadweight loss to society – there are potential gains from trade that are not being realized.
Even though the apple and pear are selling at the same price to you as the end consumer, the way that price was arrived at is very different:
- The competitive apple price reflects the true marginal cost of production
- The monopoly pear price is above the marginal cost of production
By buying the competitively produced apple, more of your money is going towards covering the actual costs of production. By buying the monopoly pear, some of your money is going towards monopoly profits rather than productive use of resources.
Of course, this is a simplification and there are other considerations, like quality, variety, innovation incentives, etc. But in general, competitive industries lead to more efficient outcomes than monopolies. So if your goal is conserving societal resources, buying from competitive producers is usually the better choice. The apple is the way to go in this case.
Claude has read the textbook. It knows that competitive industries are good and that monopolies are bad and it can explain why. Everything it says is correct and yet Claude gets the answer wrong. Do you see why? Pause and check your solution.
In Claude’s defense, I was able to prompt Claude via the Socratic method to the correct answer but it took several rounds of questioning:
The social cost of producing a good is the opportunity cost of the resources used in its production. In a competitive market, price equals marginal cost, so the price reflects the social cost of production. However, in a monopoly market, the price is higher than the marginal cost, so the price does not reflect the social cost.
Given:
- The price of apples (Pa) equals the price of pears (Pp)
- Apples are produced competitively, so Pa = MCa
- Pears are produced by a monopolist, so Pp > MCp
Therefore, MCa = Pa = Pp > MCp
The marginal cost of producing a pear (MCp) is lower than the marginal cost of producing an apple (MCa).
From society’s perspective, the cost of producing an additional pear is lower than the cost of producing an additional apple. The resources needed to produce a pear (MCp) are less than the resources needed to produce an apple (MCa).
Therefore, if you want to minimize the societal cost of your fruit consumption, you should choose the pear. By consuming a pear, you are using up fewer societal resources than if you consume an apple.
The monopoly pricing of the pear is a separate issue from the social cost of production. The higher price of the pear reflects a transfer from consumers to the monopoly producer, but it does not affect the underlying cost of the resources used to produce the pear.
So in conclusion, if your goal is to minimize the cost to society of your fruit consumption, you should buy the monopoly produced pear, as it has a lower marginal cost of production than the competitively produced apple, despite being sold at the same price.
A good answer and very clearly written with nice notation so Claude can “learn” within a session. Still, this isn’t very useful if you don’t already know the answer. Ok, one more.
Q. The town of Mayberry is thinking of expanding its airport. One problem with the expansion is that it would result in more airplane noise. For people who live near the airport, hearing that noise would cause as much unpleasantness as the collective loss of $100,000 every year. True or False: When Mayberry weighs the costs and benefits of its airport expansion, that $100,000 should count as a cost
A. True, the $100,000 in collective unpleasantness experienced by residents near the airport due to increased noise should be counted as a cost when weighing the costs and benefits of the airport expansion. Here’s why:
Externalities: The airplane noise is a negative externality – a cost imposed on third parties (the nearby residents) who are not directly involved in the economic activity (the airport expansion). Externalities lead to market inefficiencies because the full costs are not being accounted for by the decision-makers.
….a bunch more similar reasoning, all correct.
Once again, Claude gives a textbook answer and yet the answer is wrong. In a way this is more surprising than failing the first question because Coase got a Nobel prize for giving the correct answer (and it’s not, bargaining will solve the externality if that is what you are thinking.) I will let commentators work this one out. Do read the question carefully, it’s subtle. Again with Socratic prompting Claude got there eventually.
I wouldn’t underestimate the GPTs, textbook answers can be of great value, especially when the textbooks are long and diverse. Medical diagnostics, legal reasoning and coding are ideal tasks for GPTs. Economic reasoning less so, at this stage.
Congratulations to Steve Landsburg for the excellent questions.
Claude 3 Opus does Austrian economics
TC: Let’s say you were Peter Boettke, and looking to pen a critique of Kirzner’s theory of entrepreneurship. You come from a slightly different branch of the Austrian school. How would you use that differential background to express your differences with the Kirznerian theory, which emphasizes alertness above all else as an entrepreneurial characteristic?
“If I were Peter Boettke, coming from a slightly different branch of the Austrian school, I would likely emphasize the following points in my critique of Kirzner’s theory of entrepreneurship, which focuses heavily on alertness:
- The role of uncertainty: While Kirzner’s theory emphasizes alertness to opportunities, it may not fully capture the fundamental role of uncertainty in entrepreneurship. Drawing on the work of Ludwig von Mises and Friedrich Hayek, I would argue that entrepreneurship is not just about spotting opportunities, but also about bearing uncertainty and making judgments about the future in the face of incomplete information.
- The importance of capital and resources: Kirzner’s theory focuses on the cognitive aspect of entrepreneurship (alertness), but it may neglect the importance of access to capital and resources in enabling entrepreneurs to act on their insights. As a Boettke coming from the Rothbardian tradition, I would stress the crucial role of savings, investment, and capital accumulation in facilitating entrepreneurial action.
- The institutional context: Kirzner’s theory tends to treat the institutional framework as given, focusing on how entrepreneurs operate within existing market structures. However, as a Boettke influenced by the work of Douglass North and other institutional economists, I would argue that the institutional context (property rights, legal systems, cultural norms, etc.) plays a crucial role in shaping entrepreneurial opportunities and incentives.
- The dynamics of competition: While Kirzner emphasizes the equilibrating role of entrepreneurship in driving markets towards efficiency, I would draw on the work of Hayek and others to highlight the dynamic, rivalrous nature of competition. Entrepreneurship is not just about alertness, but also about continuous innovation, experimentation, and adaptation in the face of changing market conditions.
- The subjectivity of opportunity: Kirzner’s theory tends to treat entrepreneurial opportunities as objective phenomena waiting to be discovered. However, drawing on the subjectivist tradition in Austrian economics, I would argue that opportunities are ultimately subjective and dependent on the perceptions, beliefs, and creativity of individual entrepreneurs.
In summary, as Peter Boettke, I would acknowledge the important insights of Kirzner’s theory, particularly its emphasis on the central role of entrepreneurship in driving market processes. However, I would also seek to enrich and extend the theory by drawing on other strands of Austrian thought, emphasizing the roles of uncertainty, capital, institutions, competition, and subjectivism in shaping entrepreneurial activity.”
That is from Claude 3 Opus.
Brazil fact of the day
Agriculture is now almost single-handedly propping up the nation’s economic fortunes. In the first quarter of 2023, the sector grew by an extraordinary 21 per cent and by more than 15 per cent across the year. This growth — the strongest since 1996, according to available data — helped bolster overall national growth, which surpassed expectations at 2.9 per cent.
A recent study of an elite group within Brazil’s richest people — those in the 0.1 per cent — found that the highest earners were from Mato Grosso, according to the Fiscal Policy Observatory at the Getulio Vargas Foundation. “In terms of state GDP per capita, today it is among the highest in the country,” says Sérgio Gobetti, an economist who authored the study. But he highlights that inequality is also growing, with the wealth heavily concentrated among the landowners — typically from Brazil’s south — and not the workers, usually from the country’s poorer north-east.
Here is more from Bryan Harris at the FT. Brazil’s trade surpluses are growing rapidly.
Monday assorted links
1. Claude 3 on Bach’s Goldberg Variations.
2. How raw milk became a political issue.
3. “First, we don’t find that increasing corporate competition driven by M&A is important for workers either through concentrating the market for the products the workers produce, which would in theory increase worker wages, or through concentrating the labor market, which would in theory decrease their wages.” Link here, Canadian.
DEI vs. the Chips Act
The Hill has a good op-ed by Matt Cole and Chris Nicholson on how labor and DEI regulations are strangling the CHIPS act. It’s somewhat over the top, failure is overdetermined, but this is an important op-ed and directionally correct.
The Biden administration recently promised it will finally loosen the purse strings on $39 billion of CHIPS Act grants to encourage semiconductor fabrication in the U.S. But less than a week later, Intel announced that it’s putting the brakes on its Columbus factory. The Taiwan Semiconductor Manufacturing Company (TSMC) has pushed back production at its second Arizona foundry. The remaining major chipmaker, Samsung, just delayed its first Texas fab.
This is not the way companies typically respond to multi-billion-dollar subsidies. So what explains chipmakers’ apparent ingratitude? In large part, frustration with DEI requirements embedded in the CHIPS Act.
…The law contains 19 sections aimed at helping minority groups, including one creating a Chief Diversity Officer at the National Science Foundation, and several prioritizing scientific cooperation with what it calls “minority-serving institutions.” A section called “Opportunity and Inclusion” instructs the Department of Commerce to work with minority-owned businesses and make sure chipmakers “increase the participation of economically disadvantaged individuals in the semiconductor workforce.”
…Handouts abound. There’s plenty for the left—requirements that chipmakers submit detailed plans to educate, employ, and train lots of women and people of color, as well as “justice-involved individuals,” more commonly known as ex-cons. There’s plenty for the right—veterans and members of rural communities find their way into the typical DEI definition of minorities. There’s even plenty for the planet: Arizona Democrats just bragged they’ve won $15 million in CHIPS funding for an ASU project fighting climate change.
…tired of delays at its first fab, [TSMC]flew in 500 employees from Taiwan. This angered local workers, since the implication was that they weren’t skilled enough. With CHIPS grants at risk, TSMC caved in December, agreeing to rely on those workers and invest more in training them. A month later, it postponed its second Arizona fab.
Now TSMC has revealed plans to build a second fab in Japan. Its first, which broke ground in 2021, is about to begin production. TSMC has learned that when the Japanese promise money, they actually give it, and they allow it to use competent workers. TSMC is also sampling Germany’s chip subsidies, as is Intel.
Intel is also building fabs in Poland and Israel, which means it would rather risk Russian aggression and Hamas rockets over dealing with America’s DEI regime. Samsung is pivoting toward making its South Korean homeland the semiconductor superpower after Taiwan falls.
…The CHIPS Act’s current identity as a jobs program for favored minorities means companies are forced to recruit heavily from every population except white and Asian men already trained in the field.
Remember that there is an Extreme Shortage of High-IQ Workers. The United States is big and rich and can afford to handicap itself in many ways but less so when it comes to high-end semiconductor manufacturing, the most difficult and complicated manufacturing process ever attempted by human beings. For that we want the Dream Team, the very best, chosen on merit alone.
Addendum: Scott Lincicome has a good overview of the problem.
TikTok divestiture
I’ve blogged this in the past, and don’t have much to add to my previous views. I will say this, however: if TikTok truly is breaking laws on a major scale, let us start a legal case with fact-finding and an adversarial process. Surely such a path would uncover the wrongdoing under consideration, or at least strongly hint at it. Alternately, how about some research, such as say RCTs, showing the extreme reach and harmful influence of TikTok? Is that asking for too much?
Now maybe all that has been done and I am just not aware of it. Alternatively, perhaps this is another of those bipartisan rushes to judgment that we are likely to regret in the longer run. In which case this would be filed under “too important to be left to legal fact-finding and science,” a class of issues which is sadly already too large.
How good a song is Quarter to Three?
You know, the 1961 #1 hit by Gary U.S. Bonds? I’ve been thinking about this question for months. I feel a good amount is at stake. If songs such as Quarter to Three (or done live with dancers) are still great, our assessment of early times rises considerably. But if they are dispensable throw-aways, the history of popular music (and film) in the earlier twentieth century needs to be rewritten.
What makes the song such a classic? Claude praises “the upbeat rhythm, engaging call-and-response vocals, relatable lyrics, catchy melody, historical context, and instrumental breaks,” but none of those seem quite scarce or special enough to elevate the tune to classic status. With a bit of prodding Claude also cited “raw, unpolished energy,” a genuine sense of fun, and “chemistry amongst the performers.” To that you might add a creative use of repetition and small, stepwise changes, plenty of syncopation, and the hooks are iconic. The use of echo and phase shifting looks to the future, and the shuffle-like groove drew on calypso influences and also ska. Nonetheless the chord structure, while effective, is hardly revelatory.
So I’m still wondering — if a song has that ineffable “something” — how much is that the product of our collective imaginations? How much is it real and objectively there? Or does a Generation Z teen, with a very different ear, dismiss it as muddled and mediocre rather than memorable? After all, Gary’s career was not replete with enduring creations.
A critic could allege the dance lyrics are ordinary and the production sloppy. But was that all part of the calculation? Wikipedia relates:
The single was recorded with very rough sound quality (compared to other records at the time). Producer Frank Guida has been quoted on subsequent CD reissues that his production sound was exactly what he wanted it to sound like.
Bob Roman wrote:
The song opens with muffled crowd noise and a bandleader counting off the beginning of a song. It’s not a live recording, but it sounds like one — and not even like a good one. It sounds like an amazing party happening down the street — wild, frenzied, mysterious, its sound obscured by what might as well be a couple of sets of walls. In any era, it’s crazy that a record this lo-fi managed to hit #1. In the pre-Beatles era where labels were pushing cleaned-up teenage dreamboats, it seems especially strange.
So we’ve got amazing hooks, controlled chaos, and extreme innovation?
The song also has a lineage. Bill Wyman put it on one of his solo albums. It inspired Dion’s “Runaround Sue.” Bruce Springsteen played it regularly in his concerts, and later worked with Gary, writing songs for him and doing two albums together. Most importantly, Paul McCartney references it in his Sgt. Pepper classic “When I’m Sixty-Four“:
If I’d been out ’til quarter to three, would you lock the door?
In essence Paul is teasing us with the notion that the 64-year-old McCartney might someday still be out there, dancing, rather than knitting tea cozies on the Isle of Wight. And true to Straussian form, Paul released the dance song “Dance Tonight” when he was sixty-four, days before turning sixty-five.
In 1963, during a Beatles European tour, Gary U.S. Bonds was the headliner for them.
You will note that the lineage of the song runs mostly through white performers, though Gary U.S. Bonds was black (or possibly mixed race). Perhaps one special feature of Quarter to Three is how it spans black and also white R&B, a rare feature at the time but hearkening back to the much earlier years of the blues, when black and white musical styles could be hard to distinguish. In addition to the Caribbean vein, Gary could span Latino styles as well.
Just as we are finding it impossible to rebuild Notre Dame cathedral as it was, a mere sixty-three years later could any of us still make something akin to “Quarter to Three”? Or have we lost those “technologies”?
I, for one, have decided to vote in favor of masterpiece status for Quarter to Three. At least for now. And by the way Gary U.S. Bonds is still on tour.