Category: Web/Tech

Do start-ups formed during recessions fare better?

by  Tyler Cowen in

We combine novel micro data with the quasi-random timing of patent decisions over the business cycle to estimate the effects of being born in the Great Recession on innovative startups. After purging ubiquitous selection biases and sorting effects, we find that recession startups experience better long-term outcomes in terms of employment and sales growth (both driven by lower mortality) and future inventiveness. While funding conditions cannot explain differences in outcomes, a labor market channel can: recession startups are better able to retain their founding inventors and build productive R&D teams around them.

Here is the full research paper by Daniel Bias and Alexander Ljungqvist.

World War II R&D and the Takeoff of the US Innovation System

by  Tyler Cowen in

That is the article subtitle, the title is “America, Jump-Started:,” and the authors of this new AER piece are Daniel P. Gross and Bhaven N. Sampat.  Here is the abstract:

During World War II, the US government’s Office of Scientific Research and Development (OSRD) supported one of the largest public investments in applied R&D in US history. Using data on all OSRD-funded invention, we show this shock had a formative impact on the US innovation system, catalyzing technology clusters across the country, with accompanying increases in high-tech entrepreneurship and employment. These effects persist until at least the 1970s and appear to be driven by agglomerative forces and endogenous growth. In addition to creating technology clusters, wartime R&D permanently changed the trajectory of overall US innovation in the direction of OSRD-funded technologies.

This is very important work, and among other things it may help explain the productivity slowdown starting in the early 1970s (that is my speculation, not from the authors).  Recommended, for all those who follow these topics.

Here are earlier, less gated copies.

Can ChatGPT assist in picking stocks?

by  Tyler Cowen in

I don’t believe this result will hold up, but I am happy to present the opposing point of view:

This paper studies whether ChatGPT-4 with access to the internet is able to provide valuable investment advice and evaluate financial information in a timely manner. Using a live experiment, we find a positive correlation between ChatGPT-4 ratings and future earnings announcements and stock returns. We find evidence that ChatGPT-4 adjusts ratings in response to earnings surprises and news events information in a timely manner. An investment strategy based on “attractiveness ratings” by ChatGPT-4 yields positive returns.

That is from a new paper by Matthias Pelster and Joel Val.

EconGoat for Claude 2.1

by  Tyler Cowen in

Claude 2’s large context window and easy file upload feature have made this one of the best LLMs.

As of November 21, 2023, Anthropic’s Claude 2.1 has the largest context window of any of the leading LLMs. Paid users can upload and query up to around 150,000 words of text, more than enough to fit this 95,000 word book. But even the free plan allows you to analyze about 75,000 words, which in this book equates to about six chapters.

Here are more details on how to use Claude 2.1 to interrogate GOAT: Who is the Greatest Economist of All Time, and Why Does It Matter?

The Indian Challenge to Blockchains: Digital Public Goods

by  Alex Tabarrok in

In my post, Blockchains and the Opportunity of the Commons, I explored the potential of blockchains to create new commons:

Blockchains and tokenization are a way to incentivize the creation of a commons. A commons is an unowned place, platform, or protocol that helps people to meet, communicate and transact. Commons underlying modern life include TCP/IP, SMTP, HTTP, GPS and the English language. We don’t see these commons clearly because they are free, ubiquitous and, like air, taken for granted. What we do see are platforms like Airbnb, Uber and the NYSE and places to meet and communicate like OkCupid, Twitter, Facebook and YouTube. What blockchain and tokenization offer is the possibility of creating commons to replace all of these services and much more.

For the most part, the potential has not been realized. But the core idea of substituting a protocol for a firm has been taken in a different direction in India. Instead of blockchains, India has been experimenting with digital public goods. A digital public good is open source software with open data and open standards–available for use or even modification and adaption by anyone. The blockchain community, for example, has long aspired to develop a blockchain-based Uber, connecting drivers and riders without a corporate intermediary. India has achieved this through digital public goods instead.

Namma Yatri is an open-source, open-data Uber-like protocol with 100% of the commission flowing directly from rider to driver. Namma Yatri is built on the Beckn Protocol, a product of the Beckn Foundation which is backed by Infosys co-founder Nandan Nilekani (Tyler and I had the opportunity to talk with many people behind the project including Nandan on a recent trip to India). Namma Yatri has booked over 15 million trips in just one year of operation, mostly in one city, Bangalore. I expect it will expand rapidly.

Namma Yatri is only one example of a digital public good in the India Stack, a collection that includes identity (Aadhaar), payments (UPI) and digital data sharing (e.g. digital lockers). Since its launch in 2008, for example, India’s Aadhaar system has created a digital identity for over 1.2 billion people allowing them to open some 650 million bank accounts. This has enhanced financial inclusion and facilitated direct government payments of pensions and rations, reducing corruption. Likewise, the UPI system built modern payment rails which are then leveraged by banks and firms such as Google Pay and WhatsApp. The resulting payments system does some 10 billion transactions a month and is one of the fastest and lowest cost in the world.

Challenges remain. The development of digital public goods relies on funding from non-profits, governments, and private consortiums, raising questions about long-term sustainability. These goods need regular maintenance and updates, and some require backend support. Namma Yatri began as a completely free app for drivers and users but if there is a problem who do you call? To support the back-end office, and to pay for updated inputs (such as maps) the service has started to use a subscription fee. Nothing wrong with that but it’s a reminder that firms are not so easily dispensed with. Privacy is another concern. While blockchains offer privacy at the technology layer, privacy for digital public goods depend on legal and normative frameworks. For instance, India’s Aadhaar system is legally restricted from police use, a smart balance that needs to be maintained in changing times.

Despite these challenges, there is no denying that India has built digital public goods at scale in a way that demonstrates an alternative pathway for digital infrastructure and a challenge to blockchains.

Labor market evidence from ChatGPT

by  Tyler Cowen in

So far some of the main effects are quite egalitarian:

Generative Artificial Intelligence (AI) holds the potential to either complement knowledge workers by increasing their productivity or substitute them entirely. We examine the short-term effects of the recent release of the large language model (LLM), ChatGPT, on the employment outcomes of freelancers on a large online platform. We find that freelancers in highly affected occupations suffer from the introduction of generative AI, experiencing reductions in both employment and earnings. We find similar effects studying the release of other image-based, generative AI models. Exploring the heterogeneity by freelancers’ employment history, we do not find evidence that high-quality service, measured by their past performance and employment, moderates the adverse effects on employment. In fact, we find suggestive evidence that top freelancers are disproportionately affected by AI. These results suggest that in the short term generative AI reduces overall demand for knowledge workers of all types, and may have the potential to narrow gaps among workers.

That is from a new paper by Xiang Hui, Oren Reshef, and Luofeng Zhou, via Fernand Pajot.  And here is an FT summary of some key results.

I would stress this point, however.  As more ordinary life and commerce structures itself around AI, more and more AI-driven or AI-enable projects will become possible.  That will favor those who are good at conceiving of projects and executing them, and those longer-run effects may well be less egalitarian.

Words to live by

by  Tyler Cowen in

I propose a model of a social media platform which manages a two-sided market composed of content producers and consumers. The key trade-off is that consumers dislike low-quality content, but including low-quality content provides attention to producers, which boosts the supply of high-quality content in equilibrium. If the attention labor supply curve is sufficiently concave, then the platform includes some low-quality content, though a social planner would include even more.

That is from the job market paper of Karthik Srinivasan of University of Chicago Booth School of Business.  Via Gavin Leech.

Autonomous Vehicles Lower Insurance Costs

by  Alex Tabarrok in

The insurance giant Swiss RE did a study comparing human drivers with Waymo autonomous vehicles in the same zip-codes and found that autonomous vehicles generated significantly fewer insurance claims.

This study compares the safety of autonomous- and human drivers. It finds that the Waymo One autonomous service is significantly safer towards other road users than human drivers are, as measured via collision causation. The result is determined by comparing Waymo’s third party liability insurance claims data with mileage- and zip-code-calibrated Swiss Re (human driver) private passenger vehicle baselines. A liability claim is a request for compensation when someone is responsible for damage to property or injury to another person, typically following a collision. Liability claims reporting and their development is designed using insurance industry best practices to assess crash causation contribution and predict future crash contributions. In over 3.8 million miles driven without a human being behind the steering wheel in rider-only (RO) mode, the Waymo Driver incurred zero bodily injury claims in comparison with the human driver baseline of 1.11 claims per million miles (cpmm). The Waymo Driver also significantly reduced property damage claims to 0.78 cpmm in comparison with the human driver baseline of 3.26 cpmm. Similarly, in a more statistically robust dataset of over 35 million miles during autonomous testing operations (TO), the Waymo Driver, together with a human autonomous specialist behind the steering wheel monitoring the automation, also significantly reduced both bodily injury and property damage cpmm compared to the human driver baselines.

The Waymo vehicles are in San Francisco and Phoenix so this doesn’t mean that autonomous vehicles are better everywhere. Also, when we say autonomous vehicles we really mean the entire Waymo system including backup. In addition, there are some differences that are hard to account for such as human drivers use more freeways even in the same zip codes. Nevertheless, it is clear that autonomous vehicles are happening. I predict that some of my grandchildren will never learn to drive and their kids won’t be allowed to drive.

Behavioral Economics and GPT-4: From William Shakespeare to Elena Ferrante

by  Tyler Cowen in

There is a new paper on LLMs by Gabriel Abrams, here is the abstract:

We prompted GPT-4 (a large language model) to play the Dictator game, a classic behavioral economics experiment, as 148 literary fictional characters from the 17th century to the 21st century. 

Of literary interest, this paper analyzed character selfishness by century, the relative frequency of literary character personality traits, and the average valence of these traits. The paper also analyzed character gender differences in selfishness.

From an economics/AI perspective, this paper generates specific and quantifiable Turing tests which the model passed for zero price effect, lack of spitefulness and altruism, and failed for human sensitivity to relative ordinal position and price elasticity (elasticity is significantly lower than humans). Model updates from March to August 2023 had relatively minor impacts on Turing test outcomes.

There is a general and mainly monotonic decrease in selfish behavior over time in literary characters. 50% of the decisions of characters from the 17th century are selfish compared to just 19% of the decisions of characters from the 21st century. Overall, humans exhibited much more selfish behavior than AI characters, with 51% of human decisions being selfish compared to 32% of decisions made by AI characters.

Historical literary characters have a surprisingly strong net positive valence across 2,785 personality traits generated by GPT-4 (3.2X more positive than negative). However, valence varied significantly across centuries. The most positive century, in terms of personality traits, was the 21st — over 10X the ratio of positive to negative traits. The least positive century was the 17th at just 1.8X. “Empathetic,” “fair” and “selfless,” were the most overweight traits in the 20th century. Conversely, “manipulative,” “ambitious” and “ruthless” were the most overweight traits in the 17th century.

Male characters were more selfish than female characters: 35% of male decisions were selfish compared to just 24% for female characters. The skew was highest in the 17th century where selfish decisions for male and female were 62% and 20% respectively.

This analysis offers a specific and quantifiable partial Turing test. In a few ways, the model is remarkably human-like; The key human-like characteristics are the zero price effect, lack of spitefulness and altruism. However, in other ways, GPT-4 reflects unusual or inhuman preferences. The model does not appear to have human sensitivity to relative ordinal position and has significantly lower price elasticity than humans.

Model updates in GPT-4 have made it slightly more sensitive to ordinal value, but not more selfish. The model shows preference consistency across model runs for each character with respect to selfishness.

To which journal might you advise him to send this paper?