AI links, 5/25/2025

Tyler Cowen on using LLMs as a professor; Ben Thompson on AI disrupting the Web Ad model; Charlie Guo explains the disruption; Sholto Douglas on rapid AI progress

Tyler Cowen writes,

Would it be so crazy to put o3, or some other advanced AI model, on the dissertation committee, in lieu of the traditional “outside reader”?

…The most ambitious professors will learn how to use the AI models themselves, and give their students the exact same quality feedback the students will be able to get for themselves. The best among the professors will learn to be inspirational mentors, coaches, and networking connectors. They will very directly help their students get somewhere in the real world. Those are some skills the AIs cannot copy, at least not anytime soon.

He asks an LLM to come up with ideas for reconfiguring higher ed around AI, and he links to the response he got. The response is good, but is it as good as my network university idea?

Ben Thompson writes,

What was not viable in the 1990s, nor at any time since then, was something like micro-transactions for content. One obvious problem is that the fee structure of credit cards don’t allow for very small transactions; another problem is that the costs to product content are front-loaded, and the potential payoff is both back-loaded and unpredictable, making it impossible to make a living. The biggest problem of all, however, is that micro-transactions are anti-human: forcing a potential content consumer to continually decide on whether or not to pay for a piece of content is alienating, particularly when plenty of alternatives for their scarce attention exist.

As Clay Shirky put it, the “mental transaction costs” were a barrier to using micropayments to access content. As a result, the advertising model dominated the web, for better or worse.

But what happens when AI “agents” are accessing content and services, so that the agents handle the mental processing of deciding when to make a small payment?

on the agentic web native digital payments are both viable and the best possible way to not only keep the web alive, but also in the process create better and more useful AI.

…What is possible — not probable, but at least possible — is to in the long run build an entirely new marketplace for content that results in a new win-win-win equilibrium.

Thompson wants crypto to be central to this new marketplace. But I think that is wrong. Crypto is too computationally intensive. An ordinary database would provide a better ledger than a decentralized database using cryptocurrency.

Charlie Guo writes,

I'm already seeing firsthand companies driving new business via ChatGPT "referrals," with no idea how to manage or influence the AI-powered inbound. Which led to today's deep dive: how these AI search systems work, strategies for staying in their good books, and the future implications of this paradigm shift.

I remember that when I first launched homefair.com, everyone came to it via the home page. So it was important to design the home page to show people where they might want to click next. But then, around 1997 or so, with Google people were going directly to other pages on the site. So now I had to design those other pages to show people where they might want to click next. With AI, the consumer may find the desired information without ever landing on any of your pages.

Pointer from Tyler Cowen, who also points to this podcast with Sholto Douglas of Anthropic. Strongly recommended. Douglas conveys the confidence that AI researchers have that progress is accelerating rapidly. To me, this means that the social and business environment is really hard to predict, even a couple of years out from now. I continue to believe that the biggest risk you can take of making a fool of yourself is to make statements of the form “AI will never be able to ____.”

substacks referenced above:

@

SEO for AI: A look at Generative Engine Optimization

@

Everyone's Using AI To Cheat at School. That's a Good Thing.