Category: Web/Tech

ChatGPT as marketing triumph

by  Tyler Cowen in

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

Consider the name itself, ChatGPT. Many successful tech products have happy, shiny, memorable names: Instagram, Roblox, TikTok. Someone comes up with the name, perhaps in a brainstorming session, and then it is taken to the broader world. With luck, it will prove sticky and continue to invoke a warm glow.

ChatGPT, as a name, does not have a comparably simple history. Originally the service was GPT-2, then GPT-3. GPT-3.5 was then released, and ChatGPT was announced in November 2022. ChatGPT-3.5 was cumbersome, and by that point it was running the risk of becoming obsolete numerically. So ChatGPT became the popular moniker. When GPT-4 was released in March, there was the chance to switch to just GPT-4 and treat ChatGPT as a kind of interim brand, connected with the 3.5 edition, but no — ChatGPT it is.

I still prefer to call it GPT-4, to make it clear which version of the service I am using. GPT-3.5 remains available for free and is widely used, and so my inner persnickety nag feels the two should not be confused. But they are confused, and almost everyone (present company excluded) just says ChatGPT. Again, this is not textbook marketing, but users seem fine with it.

At this point, you might be wondering what that GPT stands for. It is “Generative Pre-Trained Transformer.” How many of its users could tell you that? How many of its users could explain what it means? I can imagine a marketing team sitting around a conference table insisting that GPT is too opaque and just won’t do.

But at OpenAI, software engineers and neural net experts are the dominant influences. So Americans get a name that has not been heavily market-tested and does not necessarily fit well into an advertising slogan. Maybe they like it, especially if they are sick of more and more of the economy being run by marketers.

There is more at the link.  The part of GPT-4 that comes across as “focus group tested” — the neutering of Sydney and the like — is also the part that is not entirely popular, even if it is politically wise.

Ho hum

by  Tyler Cowen in

And trained on The New Testament, I might add, a text which exists in a great number of languages.

How effective was the IAEA?

by  Tyler Cowen in

Here is the Open AI call for international regulation, most of all along the lines of the International Atomic Energy Agency.  I am not in general opposed to this approach, but I think it requires very strong bilateral supplements, from the United States of course.  Which in turn requires U.S. supremacy in the area, as was the case with nuclear weapons.  From a 564 pp. official work on the topic:

For nearly forty years after its birth in 1957 the IAEA remained essentially irrelevant to the nuclear arms race. (p.22)

There is also this:

However, in the late 1950s and early 1960s it was not the failure of the IAEA’s functions as a ‘pool’ or ‘bank’ or supplier of nuclear material that inflicted the most serious blow on the organization, on its safeguards operation and eventually on Cole himself. For a variety of reasons, the Agency’s chief patron, the USA, chose to arrange nuclear supplies bilaterally rather than through the IAEA. One reason was that the IAEA had been unable to develop an effective safeguards system. Another was that in a bilateral arrangement it was the US Administration, under the watchful eyes of Congress, that chose the bilateral partner rather than leaving the choice to an international organization that would have to respond to the needs of any Member State whatever its political system, persuasion or alliance. But the most serious setback came in 1958 when, for overriding political reasons, the USA chose the bilateral route in accepting the safeguards of EURATOM as equivalent to — in other words as an acceptable substitute for — those of the IAEA.

It is frequently suggested that the IAEA has been partially captured by the nuclear sector itself.  I do not consider that bad news, but it is a sobering thought for those expecting too much from this approach.  Do note that it took years to set up the agency, and furthermore when North Korea wanted to acquire nuclear weapons the country simply left the agency and broke its earlier agreement.  Perhaps the greatest gain from this approach is that the non-crazy nations have a systematic multilateral framework to work within, should they decide to defer to the external, bilateral pressure from the United States?

On the other side, my fear is that the international agreement will lead to excess regulation at the domestic level.

There is also this:

The fact that Iraq’s nuclear weapon programme had been under way for several years, perhaps a decade, without being detected by the IAEA, led to sharp criticism of the Agency and posed the most serious threat to the credibility of its safeguards since they had first been applied some 30 years earlier.

All of these issues could use much more intelligent discussion.

Is software eating Japan? (from my email)

by  Tyler Cowen in

I came across a great series of posts by Richard Katz about Japan and information technology. It shows that software has not eaten Japan (for now?).

Some interesting facts:– “by 2025, 60% of Japan’s large companies will be operating core systems that are more than 20 years old. Would anyone today use a 2005 PC?”

– ” It is worth noting that the OECD shows Japan suffering an actual drop in output per employee in ICT business services from 2005 through 2021, its ICT productivity having peaked out in 1999. By contrast, Korea’s productivity grew 41% in the same period”

– “Among high school boys, Japan ranks number one in using the Internet to search for information on a particular topic several times a day, and Japan’s girls come in second among OECD girls. Japanese boys also rank number one in single-player online games. On the other hand, these boys score at, or near, the bottom on other activities, such as multi-player online games or uploading their own content”.

– In an OECD study, Japan is third from the bottom when it comes to the share of students ” who foresee having a career, not just in ICT, but in any area of science or engineering” [that was quite shocking to me, to be honest. Positively shocking, on the other hand, was that Portugal leads the pack]

– “What’s even more remarkable is that Japan’s top performers on the math and science tests come in dead last in the share who foresee themselves working in science or engineering”.

– Maybe this provides a bit of an explanation: ” In 2021, the average annual income of a Japanese ICT staffer was just ¥4.38 million ($34,466), down 4% from 2019. That was 2% below the median salary in Japan, whereas in the US and China, ICT salaries are 8-10% above the median.”

sources: https://richardkatz.substack.com/p/2025-digital-cliff-part-i

https://richardkatz.substack.com/p/metis-2025-digital-cliff-part-ii

That is all from Krzysztof Tyszka-Drozdowski.

Ezra Klein and Jean Twenge on teen mental health

by  Tyler Cowen in

They did an NYT podcast, here is the transcript, here is the podcast itself.  Excerpt:

EZRA KLEIN: Violent crime actually gets at why I wanted to ask this before we got into smartphones, because I’m much more familiar with the debate over crime than I am over the debate over suicide trend lines. And the thing that has always striking to me, and I think really underplayed in our national discourse over crime, is that we don’t really understand it, that if you go into the ’80s and the ’90s, you see crime goes way, way up in the 70s, way, way up in all kinds of different jurisdictions, more or less all at the same time.

And then it begins going way, way down in all kinds of different places all at the same time. So you have stories that people know, like there’s New York with Rudy Giuliani and broken windows policing and stop and frisk. But it also goes down in all these places that didn’t do what New York City did. It’s so widespread, both the rise and the fall, that you end up having researchers trying to think of even broader explanations, like whether or not lead and the amount of lead in young kids’ bloodstreams — and thus, the effect on their executive function when they got older, maybe that’s the causal mechanism.

And it made me wonder if there isn’t a chance that suicide and teen mental health is like that, that it has this kind of all the way up, all the way down in all places, and we don’t really understand why pattern to it.

Do read/listen to the whole thing, interesting throughout.

My Conversation with Simon Johnson

by  Tyler Cowen in

Here is the audio, video, and transcript.  Here is part of the episode description:

What’s more intense than leading the IMF during a financial crisis? For Simon Johnson, it was co-authoring a book with fellow economist (and past guest) Daron Acemoglu. Written in six months, their book Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity, argues that widespread prosperity is not the natural consequence of technological progress, but instead only happens when there is a conscious effort to bend the direction and gains from technological advances away from the elite.

Tyler and Simon discuss the ideas in the book and on Simon’s earlier work on finance and banking, including at what size a US bank is small enough to fail, the future of deposit insurance, when we’ll see a central bank digital currency, his top proposal for reforming the IMF, how quickly the Industrial Revolution led to widespread prosperity, whether AI will boost wages, how he changed his mind on the Middle Ages, the key difference in outlook between him and Daron, how he thinks institutions affect growth, how to fix northern England’s economic climate, whether the UK should join NAFTA, improving science policy, the Simon Johnson production function, whether MBAs are overrated, the importance of communication, and more.

And here is one excerpt:

COWEN: If institutions are the key to economic growth, as many people have argued — Daron and yourself to varying degrees — why, then, is prospective economic growth so hard to predict?

In 1960, few people thought South Korea would be the big winner. It looked like their institutions were not that good. It was a common view: oh, Philippines, Sri Lanka — then Ceylon — would do quite well. They had English language to some extent. They seemed to have okay education. And those two nations have more or less flopped. South Korea took off. It’s now, per capita income roughly equal to France or Japan. Doesn’t that mean it’s not about institutions? Because institutions are pretty sticky.

JOHNSON: Yes, I think of institutions as being part of the hysteresis effect, if you can get it in a positive way, that if you grow and you strengthen institutions, which South Korea has done, it makes it much harder to relapse. There are plenty of countries that had spurts of growth without strong institutions and found it hard to sustain that.

You make a very good point about the early 1960s, Tyler. There wasn’t that much discussion that I’ve seen about institutions per se, but education — yes, absolutely. Culture — people made the same comparisons. They said, “Confucian culture is no good or won’t lead to growth. That’s a problem, for example, for South Korea.” That turned out to be wrong.

I think institutions are sticky. I think history matters a lot for them. They’re not predestination, though. You could absolutely carve your own way, but the carve-your-own way is harder when you start with institutions that are more problematic, less democratic, more autocratic control, less protection of property rights.

All of these things can go massively wrong, but building better institutions and making them sustainable, like Eastern Europe — the parts of the former Soviet Empire that managed to escape the Soviet influence after 1989, 1991 — I think those countries have worked long and hard, with very mixed results in some places, to build better institutions. And the EU has helped them in that regard, unquestionably.

I enjoyed this session with Simon.

Might AI possibly boost social trust?

by  Tyler Cowen in

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

The point is not that LLMs won’t be used to create propaganda — they will — but that they offer users another option to filter it out. With LLMs, users can get the degree of objectivity that they desire, at least after they learn how they work.

Still, this is a safe prediction: Within a year or two, there will be a variety of LLMs, some of them open source, and people will be able to use them to generate the kinds of answers they want. You might wonder how this is an improvement on the status quo, which now affords viewers a choice of polarized media sources such as Fox News or MSNBC, not to mention voices of all sorts on Twitter.

I hold out some hope for improvement in part because LLMs operate on a “pull” basis — that is, you ask them for what you want. Even if you are working with a “right-wing” LLM, you can always ask it for a left-wing or centrist perspective, and vice versa. It would be like watching Fox and having a button on your remote that you can click to get an opposing or contrasting view — within seconds. This is a vast improvement over cable TV, in terms of immediacy if nothing else. LLMs also make it very easy to generate a debate or a “compare and contrast” answer on just about any issue.

Again, there is no knowing how much balance people will want. But at least AI will make it easier to get balance if they so choose. That seems better than having a particular cable TV channel or a pre-established Twitter feed as the baseline default.

The impersonal nature of many LLMs may also be a force for ideological balance. Currently, many left-wingers don’t want to switch to Fox (or visit their neighbor) to hear a different perspective because they find the personalities offensive or obnoxious. LLMs offer the potential to sample points of view in their driest, least provocative and most analytically argued forms.

We’ll see how this goes, and any analysis or prediction here should be quite hedged, but I am more optimistic than many people on this front.

Open source chat, no censorship

by  Tyler Cowen in

Here it is, like it or not…

 

Generative AI and firm values

by  Tyler Cowen in

What are the effects of recent advances in Generative AI on the value of firms? Our study offers a quantitative answer to this question for U.S. publicly traded companies based on the exposures of their workforce to Generative AI. Our novel firm-level measure of workforce exposure to Generative AI is validated by data from earnings calls, and has intuitive relationships with firm and industry-level characteristics. Using Artificial Minus Human portfolios that are long firms with higher exposures and short firms with lower exposures, we show that higher-exposure firms earned excess returns that are 0.4% higher on a daily basis than returns of firms with lower exposures following the release of ChatGPT. Although this release was generally received by investors as good news for more exposed firms, there is wide variation across and within industries, consistent with the substantive disruptive potential of Generative AI technologies.

A significant effect, here is the new NBER working paper from Andrea L. Eisfeldt, Gregor Schubert, and Miao Ben Zhang.

Who are the real AI experts?

by  Tyler Cowen in

That is the topic of my 2x as long as usual Bloomberg column, here is one excerpt:

It almost goes without saying that there are different kinds of expertise. National security specialists, for example, confront dangerous risks to America all the time, and they have to develop a synthetic understanding of how to respond. How many of them have resigned from the establishment to become AI Cassandras? I haven’t seen a flood of protests, and these are people who have studied how destructive actions can amplify through a broader social and economic order. Perhaps they are used to the idea that serious risks are always with us.

And here is more:

When it comes to AI, as with many issues, people’s views are usually based on their priors, if only because they have nowhere else to turn. So I will declare mine: decentralized social systems are fairly robust; the world has survived some major technological upheavals in the past; national rivalries will always be with us (thus the need to outrace China); and intellectuals can too easily talk themselves into pending doom.

All of this leads me to the belief that the best way to create safety is by building and addressing problems along the way, sometimes even in a hurried fashion, rather than by having abstract discussions on the internet.

So I am relatively sympathetic to AI progress. I am skeptical of arguments that, if applied consistently, also would have hobbled the development of the printing press or electricity.

I also believe that intelligence is by no means the dominant factor in social affairs, and that it is multidimensional to an extreme. So even very impressive AIs probably will not possess all the requisite skills for destroying or enslaving us. We also tend to anthropomorphize non-sentient entities and to attribute hostile intent where none is present.

Many AI critics, unsurprisingly, don’t share my priors. They see coordination across future AIs as relatively simple; risk-aversion and fragility as paramount; and potentially competing intelligences as dangerous to humans. They deemphasize competition among nations, such as with China, and they have a more positive view of what AI regulation might accomplish. Some are extreme rationalists, valuing the idea of pure intelligence, and thus they see the future of AI as more threatening than I do.

So who exactly are the experts in debating which set of priors are more realistic or useful?

Recommended!

What do LLMs expect for the economy?

by  Tyler Cowen in

I introduce a survey of economic expectations formed by querying a large language model (LLM)’s expectations of various financial and macroeconomic variables based on a sample of news articles from the Wall Street Journal between 1984 and 2021. I find the resulting expectations closely match existing surveys including the Survey of Professional Forecasters (SPF), the American Association of Individual Investors, and the Duke CFO Survey. Importantly, I document that LLM based expectations match many of the deviations from full-information rational expectations exhibited in these existing survey series. The LLM’s macroeconomic expectations exhibit under-reaction commonly found in consensus SPF forecasts. Additionally, its return expectations are extrapolative, disconnected from objective measures of expected returns, and negatively correlated with future realized returns. Finally, using a sample of articles outside of the LLM’s training period I find that the correlation with existing survey measures persists – indicating these results do not reflect memorization but generalization on the part of the LLM. My results provide evidence for the potential of LLMs to help us better understand human beliefs and navigate possible models of nonrational expectations.

That is from a new paper by J. Leland Bybee, via Paul Goldsmith-Pinkham.

My Conversation with the excellent Kevin Kelly

by  Tyler Cowen in

Here is the audio, video, and transcript.  Here is part of the summary:

…Kevin and Tyler start this conversation on advice: what kinds of advice Kevin was afraid to give, his worst advice, how to get better at following advice, and whether people who ask for advice really want it in the first place. Then they move on to the best places to see traditional cultures in Asia, the one thing in Kevin’s travel kit he can’t be without, his favorite part of India, why he’s so excited about brain-computer interfaces, how AI will change religion, what the Amish can teach us about tech adoption, the most underrated documentary, his initial entry point into tech, why he’s impressed by the way Jeff Bezos handles power, the last thing he’s changed his mind about, how growing up in Westfield, New Jersey affected him, his next project called the Hundred Year Desirable Future, and more.

Here is one excerpt:

COWEN: Do you ever feel that if you don’t photograph a place, you haven’t really been there? Does it hold a different status? Like you haven’t organized the information; it’s just out on Pluto somewhere?

KELLY: Yes, I did. When I was younger, I had a religious conversion and I decided to ride my bicycle across the US. And part of that problem — part of the thing was that I was on my way to die, and I decided to leave my camera behind for this magnificent journey of a bicycle crossing the US. It was the most difficult thing I ever did, because I was just imagining all the magnificent pictures that I could take, that I wasn’t going to take. I took a sketchbook instead, and that appeased some of my desire to capture things visually.

But you’re absolutely right. It was a little bit of an addiction, where the framing of a photograph was how I saw the world. Still images: I was basically, in my head, clicking — I was clicking the shutter at the right moments when something would happen. That, I think, was not necessarily healthy — to be so dependent on that framing to enjoy the world.

I’ve learned to wean myself off from that necessity. Now I can travel with just a phone for the selfies that you might want to take.

COWEN: Maybe the earlier habit was better.

Recommended.  And here is Kevin’s new book Excellent Advice: Wisdom I Wish I’d Known Earlier.

How to talk on Zoom

by  Tyler Cowen in

In a study last year, people who were face-to-face responded to yes/no questions in 297 milliseconds, on average, while those on Zoom chats took 976 milliseconds. Conversational turns — handing the mic back and forth between speakers, as it were — exhibited similar delays. The researchers hypothesized that something about the scant 30- to 70-millisecond delay in Zoom audio disrupts whatever neural mechanisms we meatbags use to get in sync with one another, that magic that creates true dialogue.

And the data:

The result is the largest-ever database of one-on-one Zoom conversations. It’s called CANDOR, short for Conversation: A Naturalistic Dataset of Online Recordings. Reece and his colleagues examined more than 1,600 conversations — some 850 hours and 7 million words total. The researchers paired volunteers, people who had never met each other, and asked them to hop on Zoom for half an hour about any old thing — with Record turned on. Which means that unlike most conversational databases, CANDOR didn’t just encode their words, which were transcribed automatically, by digital algorithms. It also automatically captured things like the tone, volume, and intensity of conversational exchanges, recording everything from facial expressions to head nods to the number of “ums” and “yeahs.”

And some results:

But loudness, it turns out, isn’t as good a metric as intensity — maybe because intensity is more subtle, a combination of the frequencies and sibilance of speech and the emotion conveyed by everything from tone to body language. To help the computer to assess something so ineffable — like, what is this thing you humans call love? — the CANDOR team fed it the Ryerson Audio-Visual Database of Emotional Speech and Song. That enabled the candorbots to draw on more than 7,000 recordings of 24 actors saying and singing things with different emotional shading, from happy or sad to fearful or disgusted. The machine found that women rated as better Zoom conversationalists tended to be more intense. The differences among men, strangely, were statistically insignificant. (The reverse was true for happiness. Male speakers who appeared to be happier were rated as better conversationalists, while the stats for women didn’t budge.)

Then there’s nodding. Better-rated conversationalists nodded “yes” 4% more often and shook their heads “no” 3% more often. They were not “merely cheerful listeners who nod supportively,” the researchers note, but were instead making “judicious use of nonverbal negations.” Translation: An honest and well-timed no will score you more points than an insincere yes. Good conversationalists are those who appear more engaged in what their partners are saying.

Here is the full Adam Rogers article, and most notably the participants did not dislike talking on Zoom per se.  Via the excellent Samir Varma.