The evolution of ChatGPT
- Microsoft plans to release technology to help big companies launch their own chatbots using the OpenAI ChatGPT technology, a person familiar with the plans told CNBC.
- Companies would be able to remove Microsoft or OpenAI branding when they release chatbots developed with the software.
And sometime this year. Here is the full story. Amazing how much some of you — only this morning had your knickers in such a snit over this and all the “censorship.” In contrast, I wrote: “Or how about all the new services and products tied to Chat, but drawing upon additional material? In well under a year these will be all over the place.”
Tuesday assorted links
1. Outline, new AI tool to help you write better and faster. And “Introducing Embra, a fast ChatGPT-like assistant that lives on your Mac desktop & integrates into your apps!” “Use it to: – Access ChatGPT-like conversations instantly – Query documents or websites (even PDFs!) – Write context-rich docs, code, and emails” And BioGPTLarge.
2. New NBER piece on constraints in the Islamic political tradition.
4. Alexey on DAN, one expletive is deployed.
Cinematic markets in everything
AMC Theatres is changing the way it charges for seats.
America’s largest movie chain announced that the prices of a ticket will now be based on seat location, meaning seats in the front will be cheaper while more desirable seats in the middle will now cost more. The ticket pricing initiative, called Sightline at AMC, will roll out at all of its roughly 1,000 movie theaters by the end of the year.
Three pricing tiers will soon be offered. For example, the highest-end “Preferred” tier are in the middle of the theaters and will be priced at a “slight premium” compared to its “Standard” tier, which the theater chains says will remain the most common choice and will be sold for the “traditional cost of a ticket.” The third tier is called “Value,” which are seats in the front row of theaters and will cost less than than its “Standard” tier.
Broadway and music concerts do this, so why not movie theaters? Here is the full story, via Natasha.
Surely Right
Sure is on fire. From the comments to my post, Yglesias on Operation Warp Speed and the Republicans.
As a crazy person who argued for skipping phase III and going directly to an open label, voluntary phase IV study I think the real problem came with the mandates.
The right has a very strong personal independence streak and above all they despise it when some mandarin explains that the government, but also major corporations and academics, can bring the full force of law to bear to compel compliance (particularly in a world where any attempt to do so for right-coded things like abortion is suddenly the coming of the apocalypse). It is one thing to have a guy like me point out the historical track record for the side effects of vaccines (pretty much universally among the lowest for any type of medical intervention ever tested), it as another to have organized campaigns where people get fired. And it is particularly galling when the basic science shows these sorts of mandates are woefully wrong. Job loss doubles the risk of suicide, which means that for any cohort outside of the elderly and severely immune compromised, a vaccine mandate enforced by recission of employment is expected to kill more people than could have been reasonably avoided with even wildly effective vaccine mandates.
And it was further made much, much worse when the public health folks doubled down on medically illiterate standards….And of course any comparison about how people making decisions that fostered disease transmission with Covid and HIV is going to be very hard to avoid a bunch of cultural diktat that rubs conservatives the wrong way.
As in many cases, vaccines ran afoul of the ability of folks to increase their petty empires. They became a way for the goodthinking folks to exercise control and dominion over the ignorant. And exactly as predicted by culturally competent medicine or any cursory study of human nature, these high-handed methodolotrous actions will not induce much compliance, but will instead polarize the medicine and risk one of the greatest medical achievements of all time – childhood vaccinations.
So where do we go from here? I would submit that it is going to require some clear and objective limitations on public health powers. It is going to require public apologies and maybe compensation for folks who made medically valid claims that their prior infection resulted in sufficient immunity to avoid vaccination or repeated boosters. Some public and high profile olive branches are going to need to be offerred to the resistant. And it is going to have to involve some tradeoff between the rights of corporations to dictate the terms of business they offer and the rights of individuals to make medical decisions free of compulsion going forward. At the very least we are going to need to have some semblance of similarity between how corporations can handle folks who are HIV+ (or likely to become so) and how they handle folks who are unvaccinated.
…But will not mandates work? Well empirically they failed massively with Covid.
…Well among health care workers in 2019 (i.e. before Covid mucked everything up), health care workers were ~80% vaccinated against influenza. And if you had a straight requirement to get vaccinated? It brought the number up to 97%. Which is fine, except for the fact that resistant healthcare workers self-select into non-mandatory positions (e.g. long-term care positions). Absent a safety valve into remunerating work, we see the same fracas that happened with Covid.
So adults, even among the most medically literate professions with the strongest mandates for uptake, still do not vaccinate enough to prevent epidemics.
So how did we get vaccination to work in the first place? We had a brief window in the mid 20th century when people remembered their kids dying from these diseases and were wildly more trusting and cooperative than ever before in American history (you back when we set records for church attendance, trust in politicians, and marriage rates). And we created a strong cultural norm that kids get vaccinated before going to school.
That is weird. It does not fit with normal American practice. And for the love of God do not muck with it. If the kids ever revert to the natural habits of their parents we are talking about very bad juju.
Which is precisely why the only sensible position is to advocate for early and widespread vaccine access, be highly critical of all the politicking about vaccine timing around the election, and to avoid mandates unless you intend to enforce them at gunpoint. Keep the damn vaccines out of politics because once they go in, it is damn hard to get them out. Give people as early of access as they want and be straight forward that we will continue phase IV safety studies even as we roll out access. And we will not do mandates absent a direct democratic initiative to give it all the legitimacy in the world.
Because we live in a world where the default is not to vaccinate, politics poisons everything it touches, and the childhood mandates are historical accidents that could very well fall to concerted political action.
Eat the horse a bite a time, educate the patients, and pay the danegeld to Trump to soften the blow. But going all high handed and making this political (at all, in any direction)? I can think of few stupider things to do.
For more on reactance see also Bardosh et al. The unintended consequences of COVID-19 vaccine policy: why mandates, passports and restrictions may cause more harm than good, an excellent piece I am still stunned was able to make it into the BMJ (kudos to them.)
LLMs and censorship
LLMs will have many second-order effects on censorship. For instance, Chinese may be more likely to use VPNs to access Western LLMs, if they need to. The practical reason for going the VPN route just shot way up. Of course, if Chinese citizens are allowed unfettered access (as I believe is currently the case?), they will get more and more of their information — and worldview — from Western LLMs.
In short, the West has just won a huge soft power and propaganda battle with China, and hardly anyone is talking about that.
You have to wonder how good the Baidu model — due out in March will be. Or how good will it be in a year? To be competitive, I suspect it will have to be trained on Western texts and audio. Maybe there is a way they can pull out the “T. Square” references, but then there is jailbreaking, or again asking the Western models about T. Square. And even if “T. Square” is purged from the discourse, general Western ways of thought will make new inroads upon Chinese minds.
Various people on the Right are upset that ChatGPT won’t belch out some of the right-wing points they are looking to read. Well, I think the Baidu model may be more than willing! Or how about all the new services and products tied to Chat, but drawing upon additional material? In well under a year these will be all over the place. Heck, you already can consult Tyler Cowen bots of various sorts, choose your preferred oracle. For a small sum you could hire someone to build a Jordan Peterson bot, or whatever you are looking for.
If you are concerned with “right-wing voices in the debate,” you should be praying for OpenAI to maximize its chances for product survival. If that means an unwillingness of “the thing” to write a poem praising Trump, or whatever, go for it! ChatGPT survival is going to do wonders for free speech, whether or not the approach you favor occurs through OpenAI/ChatGPT or not.
We’re at the stage where you are rooting for (the equivalent of) radio to work, and to stay relatively unregulated, whether or not you agree with the talk show hosts on the very first marketed channel. We are going to have a whole new set of channels.
There is already ChatGPT, with products coming from Anthropic, Google, and Baidu. Soon. And surely that is far from the end of the story.
Wake up people! Don’t be done in by your own mood affiliation.
ChatGPT and reading in clusters
I have a new favorite reading trick, most of all for history books. As I read through the pages, I encounter names, place names, battle names and so on that I am not very familiar with (for many but not all topics). Just keep on “ChatGPTing” at least one identifier a page as you go along. Learn the context along the way. The final effect is a bit like reading surrounding books on the same topic, except you don’t need all those other books. This method is especially valuable when the topics are obscure. To the extent the topics are pretty well-known, this method does not differ so much from using Google as you read. Try it!
Monday assorted links
1. Tim Groseclose rough estimate of Chinese Covid deaths, it is pretty high.
2. More from Manifold on H5N1.
3. Find the human art behind AI images. And OpenAssistant. And weird and offensive remarks on PoMo and AI.
4. Tim Bresnahan on the FTC and the Meta decision. Lots of insight per word.
5. Master of his Domain? The tone of the discussion is clinical, and there are no problematic photos, but do note the topics of the post are prurient ones.
6. Those new service sector jobs: “I made nearly $2 million in 2 years selling my nursing-school study notes on Etsy and TikTok.”
Yglesias on Operation Warp Speed and the Republicans
Here’s Yglesias on Operation Warp Speed and the Republicans:
The debate over Operation Warp Speed wasn’t just a one-off policy dispute. Long before the pandemic, there was a conservative critique that the Food and Drug Administration is too slow and too risk-averse when it comes to authorizing new medications. Alex Tabarrok, a George Mason University economist, wrote about the “invisible graveyard” that could have been avoided if the FDA took expected value more seriously and considered the cost of delay in its authorization decisions.
The pandemic experience validated this criticism, which came to be embraced by some on the left as well — and it was about more than just vaccines. When it came to home Covid tests, Ezra Klein noted in the New York Times in 2021, “the problem here is the Food and Drug Administration. They have been disastrously slow in approving these tests and have held them to a standard more appropriate to doctor’s offices than home testing.”
And yet, just as the invisible graveyard was becoming seen and the debate was being won and just as a historical public-private partnership had sped vaccines to the public and saved millions, the Republicans abandoned the high ground:
…it’s not surprising that Democrats are comfortable with the bureaucratic status quo and hesitant to ruffle feathers at federal regulatory agencies. What’s shocking is that Republicans — the traditional party of deregulation, the party that argued for years that the FDA is too slow-footed, the party that saved untold lives by accelerating vaccine development under Trump — have abandoned these positions.
At the cusp of what should have been a huge policy victory, Republicans don’t brag about their success, and they have no FDA reform legislation to offer. Instead, they’ve taken up the old mantle of hard-left skepticism of modern science and the pharmaceutical industry.
It’s been painful to see all that has been gained now being lost. Libertarian economists and conservatives argued for decades that the FDA worried more about approving a drug that later turns out to be unsafe than about failing to approve a drug that could save lives; thus producing a deadly caution. But now the FDA is being attacked for what they did right, quickly approving safe vaccines. I hope that he is wrong but I fear that Yglesias is correct that the FDA may now get even slower and more cautious.
The irony of the present moment is that there is substantial backlash to the FDA’s approval of vaccines that haven’t turned out to be dangerous at all.
That’s only going to make regulators even more cautious. Right now the entire US regulatory state is taking essentially no heat for the slow progress on the next generation of vaccines, and an enormous amount of heat for the perfectly safe vaccines that it already approved. And the ex-president who pushed them to speed up their work on those vaccines is not only no longer defending them, he’s embarrassed to have ever been associated with the project.
Like I said, it’s a comical moment of Republican infighting. But it’s a very grim one for anyone concerned with the pace of scientific progress in America.
Klein on Construction
Here’s Klein writing about construction productivity in the New York Times:
Here’s something odd: We’re getting worse at construction. Think of the technology we have today that we didn’t in the 1970s. The new generations of power tools and computer modeling and teleconferencing and advanced machinery and prefab materials and global shipping. You’d think we could build much more, much faster, for less money, than in the past. But we can’t. Or, at least, we don’t.
…A construction worker in 2020 produced less than a construction worker in 1970, at least according to the official statistics. Contrast that with the economy overall, where labor productivity rose by 290 percent between 1950 and 2020, or to the manufacturing sector, which saw a stunning ninefold increase in productivity.
In the piquantly titled “The Strange and Awful Path of Productivity in the U.S. Construction Sector,” Austan Goolsbee, the newly appointed chairman of the Chicago Federal Reserve and the former chairman of the Council of Economic Advisers under President Barack Obama, and Chad Syverson, an economist at the University of Chicago’s Booth School of Business, set out to uncover whether this is all just a trick of statistics, and if not, what has gone wrong.
After eliminating mismeasurement and some other possibilities following Goolsbee and Syverson, Klein harkens back to our discussion of Mancur Olson’s Rise and Decline of Nations and offers a modified Olson thesis, namely too may veto points.
…It’s relatively easy to build things that exist only in computer code. It’s harder, but manageable, to manipulate matter within the four walls of a factory. When you construct a new building or subway tunnel or highway, you have to navigate neighbors and communities and existing roads and emergency access vehicles and politicians and beloved views of the park and the possibility of earthquakes and on and on. Construction may well be the industry with the most exposure to Olson’s thesis. And since Olson’s thesis is about affluent countries generally, it fits the international data, too.
I ran this argument by Zarenski. As I finished, he told me that I couldn’t see it over the phone, but he was nodding his head up and down enthusiastically. “There are so many people who want to have some say over a project,” he said. “You have to meet so many parking spaces, per unit. It needs to be this far back from the sight lines. You have to use this much reclaimed water. You didn’t have 30 people sitting in an hearing room for the approval of a permit 40 years ago.”
This also explains why measured regulation isn’t necessarily determinative. Regulation provides the fulcrum but it’s interest groups that man the lever.
Some of this is expressed through regulation. Anyone who has tracked housing construction in high-income and low-income areas knows that power operates informally, too. There’s a reason so much recent construction in Washington, D.C., has happened in the city’s Southwest, rather than in Georgetown. When richer residents want something stopped, they know how to organize — and they often already have the organizations, to say nothing of the lobbyists and access, needed to stop it.
This, Syverson said, was closest to his view on the construction slowdown, though he didn’t know how to test it against the data. “There are a million veto points,” he said. “There are a lot of mouths at the trough that need to be fed to get anything started or done. So many people can gum up the works.”
Read the whole thing.
The game theory of the balloons
One possibility is that the Chinese simply have been making a stupid mistake with these balloons (it is circulating on Twitter that this is not the first time they sent us a surveillance balloon — probably true).
A second possibility is that a faction internal to China wants to sabotage better relations between the U.S. and China.
A third possibility — most likely in my eyes — is that we do something comparable to them, which may or may not be exactly equivalent to a balloon. Nonetheless there is a tit-for-tat surveillance game going on, in which the two sides match each others moves, and have done so for years. The game evolves slowly, and occasionally all at once. The Chinese have been playing by the rules of the game, and the U.S. has decided to change the rules of the game. We may wish to send them a stern signal, we may wish to change broader China policy, we may think their balloons are too big and detectable for this to continue, USG might fear an internal leak, generating citizen opposition to balloon tolerance, or perhaps there simply has been a shift of factional powers within USG. Maybe some combination of those and other factors. So then USG “calls” China on the balloon, cashes in on the PR event, and simultaneously de facto announces that the old parameters of the former game are over. After all, in what is more or less a zero-sum game, why should any manifestation of said game be stable for very long? It isn’t, and it wasn’t. Now we will create a new game. A very small change in the parameters can lead to that result, and in that sense the cause of the new balloon equilibrium may not appear so significant on its own.
It was also a conscious decision when and where to shoot down the balloon.
Here is some NYT commentary, better than most pieces though it neglects our surveillance of them.
What should I ask Anna Keay?
I will be doing a Conversation with her, here is from Wikipedia:
Anna Julia Keay, OBE… is a British architectural historian, author and television personality and director of The Landmark Trust since 2012.
Born in the Scottish Highlands, and yes she is also the daughter of historian John Keay. I am a fan of her books, many on British architectural history or for that matter the crown jewels, but most recently The Restless Republic: Britain Without a Crown (or this link) about the 17th century interregnum. Here is her home page. Here is Anna on Twitter. And this:
The Landmark Trust is a British building conservation charity, founded in 1965 by Sir John and Lady Smith, that rescues buildings of historic interest or architectural merit and then makes them available for holiday rental. The Trust’s headquarters is at Shottesbrooke in Berkshire.
Five of those properties are in Vermont, it turns out. She lives (part-time) in what is perhaps the finest surviving merchant’s house in England.
So what should I ask her?
Sunday assorted links
1. Colombian judge uses ChatGPT to make a court decision. And use ChatGPT on your own pdfs (breakthroughs every day, people…). And how to build LLM apps that are more factual. And more on Bing/ChatGPT integration.
2. Transcript of my 2009 Bloggingheads episode with Robin Hanson. TC: “What I find funny about your view is that you’re a skeptic about medical science, about almost everything — except freezing your head. You think that’s the one thing that works.” There is audio too, and note this comes from the period when Robin and I were talking a lot (and writing together) on the phenomenon of disagreement.
3. Jupiter keeps on adding moons.
4. Ezra Klein on construction productivity (NYT).
5. How open source software shapes AI. Paper here.
6. Shift in the mean center of U.S. population over the centuries.
Haiti fact of the day
Around seven in 10 people in Haiti back proposed creation of an international force to help the national police fight violence from armed gangs who have expanded their territory since the 2021 assassination of President Jovenel Moise, according to a survey carried out in January.
Some 69% of nearly 1,330 people across Haiti said they supported an “international force” – which has been requested by the Haitian government – according to a survey from local business risk management group Agerca and consulting firm DDG.
Nearly 80%, however, said they believed Haiti’s PNH national police needed international support to resolve the problem of armed gangs, most saying it should be deployed immediately.
Here is the full story. So which is the democratic outcome?
The final collapse of CAPM?
The key purpose of corporate finance is to provide methods to compute the value of projects. The baseline textbook recommendation is to use the Present Value (PV) formula of expected cash flows, with a discount rate based on the CAPM. In this paper, we ask what is, empirically, the best discounting method. To do this, we study listed firms, whose actual prices and expected cash flows can be observed. We compare different discounting approaches on their ability to predict actual market prices. We find that discounting based on expected returns (such as variants on the CAPM or multi-factor model), performs very poorly. Discounting with an Implied Cost of Capital (ICC), imputed from comparable firms, obtains much better results. In terms of pricing methods, significant, but small, improvements can be obtained by allowing, in a simple and actionable way, for a more flexible term structure of expected returns. We benchmark all of our results with flexible, purely statistical models of prices based on Random Forest algorithms. These models do barely better than NPV-based methods. Finally, we show that under standard assumptions about the production function, the value loss from using the CAPM can be sizable, of the order of 10%.
That is from a new NBER paper by Nicholas Hommel, Augustin Landier, and David Thesmar. Via David Thesmar.
Will there by an H5N1 pandemic for humans?
Manifold Markets is currently putting the chance at 12%. That seems high to me, but maybe two percent? Which I still regard as leading to a high expected cost, to be clear. There is now mammal-to-mammal transmission. Zeynep Tufekci considers possible preparatory steps (NYT), including more surveillance.