Category: Web/Tech

Agustin Lebron on the new AI Executive Order (from my email)

by  Tyler Cowen in

Even though it’s not a compute cap (just a requirement to register, for now), surely this emboldens competitors to OpenAI/etc? If I think I can build a system up to the cap, then my playing field just got more level.

This will also spur lots of research into sample efficiency again, which IMO is a very good thing. It’s been a research area that has languished for years under the “just throw more flops at it” regime.

Finally, I think the order perversely legitimizes capabilities research by showing the world “Well, the US gov’t thinks capabilities are SO important they’re making orders about it”. More attention, more dollars, more progress.

Overall this is accelerative for AI/AGI IMO.

*Natural Selection of Artificial Intelligence∗

by  Tyler Cowen in

A new research paper:

We study the AI control problem in the context of decentralized economic production. Profit-maximizing firms employ artificial intelligence to automate aspects of production. This creates a feedback loop whereby AI is instrumental in the production and promotion of AI itself. Just as with natural selection of organic species this introduces a new threat whereby machines programmed to distort production in favor of machines can displace those machines aligned with efficient production. We examine the extent to which competitive market forces can serve their traditional efficiency-aligning role in the face of this new threat. Our analysis highlights the crucial role of AI transparency. When AI systems lack perfect transparency self-promoting machines destabilize any efficient allocation. The only stable competitive equilibrium distorts consumption down to catastrophic levels.

By Jeffrey C. Ely and Balazs Szentes.  Whether or not you agree with their approach and conclusions, we finally have a model of some of these claims.  If you are curious about possible responses, one modification might be to relax the assumption of constant returns to scale.  Rising costs will make it harder for effective, world-altering machines (as opposed to “introverted” machines) to simply keep on reproducing themselves.  Another modification would be to introduce a richer menu of principal-agent contracts between humans and machines.  As I understand the current draft, the only human strategy is “destroy the mutant machine, if detected.”  Yet if the machines are risk-neutral (are they?), an optimal principal-agent contract should be available.  Yet another modification would be to consider mutant machines that reproduce at the expense of other (heterogeneous) machines, rather than at the expense of humans; heterogeneity of production inputs might ease the way toward this conclusion.

Expensive markets in everything

by  Tyler Cowen in

…testimony in the trial revealed that Google spent a total of $26.3 billion in 2021 to be the default search engine in multiple browsers, phones, and platforms.

That number, the sum total of all of Google’s search distribution deals, came out during the Justice Department’s cross-examination of Google’s search head, Prabhakar Raghavan.

$18 billion of that goes to Apple.  Here is the full story.

*GOAT: Who is the Greatest Economist of all Time, and Why Does it Matter?*

by  Tyler Cowen in

I am pleased to announce and present my new project, available here, free of charge.  It is derived from a 100,000 word manuscript, entirely written by me, and is well described by the title of this blog post.

I believe this is the first major work published in GPT-4, Claude 2, and some other services to come.  I call it a generative book.  From the project’s home page:

Do you yearn for something more than a book? And yet still love books? How about a book you can query, and it will answer away to your heart’s content? How about a book that will create its own content, on demand, or allow you to rewrite it? A book that will tell you why it is (sometimes) wrong?

To be clear, if you’re not into generative AI, you can just download the work onto your Kindle, print it out, or read it on a computer screen.  Yet I hope you do more:

One easy place to start is with our own chatbot using GPT-4, and we’ll soon provide custom apps using Claude 2 and Llama 2. In the meantime we’ve provided instructions for how to experiment with them yourself.

Each service has different strengths and you should try more than one. You’ll see the very best performance by working with individual chapters using your own subscription to ChatGPT, Claude, or a similar service. The chapters can be read independently and in any order. Ask the AI if you’re lacking context. Try these sample questions to start.

You can ask it to summarize, ask it for more context, ask for a multiple choice exam on the contents, make an illustrated book out of a chapter, or ask it where I am totally wrong in my views.  You could try starting with these sample questions.  The limits are up to you.

Here is the Table of Contents:

1. Introduction

2. Milton Friedman

3. John Maynard Keynes

4. Friedrich A. Hayek

5. Those who did not make the short list: Marshall, Samuelson, Arrow, Becker, and Schumpeter

6. John Stuart Mill

7. Thomas Robert Malthus

8. Adam Smith

9. The winner(s): so who is the greatest economist of all time?

The site address is an easy to remember econgoat.ai.  And as you will see from the opening chapter, it is not only about economics, it is also a very personal book about me.  No Straussian here, I tell you exactly what I think, including of my personal meetings with Friedman and Hayek.  If, however, you are looking for a Straussian reading of this project — which I would disavow — it is that I am sacrificing “what would have been a normal book” to the AI gods to win their favor.

And apologies in advance for any imperfections in the technology — generative books can only get better.

Recommended.

Will AI cause a market crash/

by  Tyler Cowen in

Gary Gensler has been worrying about this, and calling for extra regulation, but the arguments behind his view are not so strong.  Here is one part from my latest Bloomberg column:

One fear is that a small number of AI base models could lead investors to herd behavior, where many of them sell (or buy) at the same time because their models have told them to. But the number of base models is likely to rise over time, not fall. AI is in a period of considerable innovation, with many startups being founded and many new trading and investing techniques being developed. Diversity, not uniformity, will reign.

The incentives of a trading firm are not to use the same model as everyone else, as that could lead them to sell into falling market panics or buy into temporarily rising prices — which is precisely what they should not do. Instead, a top trading firm will try to develop better models than its competitors. If a firm discovers that competitors are using a common model in a predictable way, it can identify the weaknesses of that model and trade against those firms.

There is much more at the link.

It is better that Elon bought Twitter

by  Tyler Cowen in

I have disagreed with most of his design decisions, do not like the name change or rebrand, and I have been disappointed by many of his tweets and points of view, often disagreeing vehemently.  That said, allowing the videos to be seen on Twitter is the right decision, and it is a very, very important decision.

So I end up glad that he bought Twitter.  I also very much like the general feed and also the “Community Notes” features, the latter supported but not initiated by Elon.

I am not sure how widely acknowledged this will be, but someone should say it, and I am happy to be the one.  In general, more attention needs to be paid to “getting one big thing right.”

How will AI remake the rules of international trade?

by  Tyler Cowen in

That is the topic of my latest Bloomberg column, here is one excerpt:

Posting a query to ChatGPT consumes a lot of energy, by one estimate 10 times more than a Google search. Currently large language models are sufficiently limited that this is not a major factor in aggregate energy consumption. But as use of AI services increases, the energy burden will rise. Countries with expensive energy, or which will not allow energy consumption to rise much for climate or regulatory reasons, will look to import their AI services from energy-rich countries.

In the future, energy-rich regions may include Spain and Morocco with solar power, South Korea with affordable nuclear power, and whichever nations are pioneers in nuclear fusion. Those nations may end up as major exporters of AI-generated data. They might draw their AI inputs from the US, but specialize in cheap calculation and information transmission. And some regions of America may join this list as well, especially if they are well-suited for solar and hydroelectric power.

To be clear, the US will export a lot of AI services, through such companies as OpenAI, Google, Meta and Anthropic. But the US is not as good at building affordable infrastructure, and that will put it at a disadvantage in the AI revolution and distribute many of the gains abroad.

It remains to be seen whether there are higher profits in selling the original source code or the more derivative electricity-driven, infrastructure-based AI calculations. Nonetheless, this is a potential economic and national security risk for the US. It could end up with a strong lead in the source product, but fall badly behind in making (“manufacturing,” you could say) the final AI outputs.

Recommended.

Obsolescence Rents

by  Tyler Cowen in

The subtitle of this new NBER working paper is: Teamsters, Truckers, and Impending Innovations.  Here is the abstract:

We consider large, permanent shocks to individual occupations whose arrival date is uncertain. We are motivated by the advent of self-driving trucks, which will dramatically reduce demand for truck drivers. Using a bare-bones overlapping generations model, we examine an occupation facing obsolescence. We show that workers must be compensated to enter the occupation – receiving what we dub obsolescence rents – with fewer and older workers remaining in the occupation. We investigate the market for teamsters at the dawn of the automotive truck as an á propos parallel to truckers themselves, as self-driving trucks crest the horizon. As widespread adoption of trucks drew nearer, the number of teamsters fell, the occupation became ‘grayer’, and teamster wages rose, as predicted by the model.

That is by Costas Cavounidis, Qingyuan Chai, Kevin Lang, and Raghav Malhotra.

Will misinformation be an electoral problem?

by  Tyler Cowen in

That is the topic of my latest Bloomberg column, here is one bit:

…consider the story that former President Barack Obama was not born in the US. It did not take off because someone forged a copy of an Indonesian birth certificate. Instead, many people approached the issue wanting to believe that Obama was not “a real American,” some dangerous tidbits were thrown their way, and off they went. The release of Obama’s US birth certificate did not convince them they were wrong.

Lies, misunderstandings, instances of self-deception: They have long been in excess supply. Blame China, Russia, social media, regular media, whomever. A potentially gullible person is already flooded with more lies in a single day than he or she can possibly evaluate.

A greater number of falsehoods just won’t matter that much — because the scarce resources are attention and focality on the demand side. How much is someone looking to believe they have been wronged? How much do they resent “the establishment”? What kinds of grudges do they hold, and against whom or what? And how well can they coordinate with others of like mind, thereby forming a kind of misinformation affinity group?

There is much more at the link.

The Productivity of Online Education Increases

by  Alex Tabarrok in

Teaching is a labor-intensive service industry for which it is difficult to increase productivity. Thus, the price of education rises over time, the Baumol effect. One of the reasons Tyler and I have put a lot of effort into online education is that it ties education to high productivity growth industries such as software and technology. Thus in the Industrial Organization of Online Education we argued:

…as more of the value of a course comes from software and less from live teaching, productivity will improve, thus removing the cost disease.

Here’s a case in point:

This is just a test but all of the videos for our textbook, Modern Principles, and our free online platform Marginal Revolution University are already subtitled in many languages and soon we will see more translations like the one above. Amazing.