Iain Schmitt

---

Manifest 2026 Recap

For the second time this year, I made it out to the Bay Area for a conference. In a write-up on my March visit for JavaOne, I wrote the following:

...as Byrne Hobart points out, by clustering a lot of people in a small location for a couple of days you end up creating a miniature industry-specialised city with all the implied nonlinear benefits.

It was a mistake to write that, not because it was wrong, but because I should have saved it for this post: the linked article is Hobart's 'Conferences as Industry Cluster Economics, Squared', which detailed his experience at Manifest 2024, a conference on prediction markets and related topics.¹ That year was my first year at Manifest, and I've gone every year since. As always, Lighthaven was the Manifest venue; rather than a cookie-cutter conference centre, it is five detached buildings on one campus with plenty of great spaces for side conversations between them.

Manifest has become a highlight of my year, and every conference I meet more attendees that I look forward to seeing again next June. Most people who are interested in prediction markets have a lot of overlap in blog readership, so when talking with people at Manifest I can use phrases like conflict/mistake theory or the Baumol effect without needing to explain myself. This means you can pack a lot of detail into the conversations you have over lunch and between talks.

This list won't do this incredible weekend justice, but here is a recap in no particular order.

1. I should have realised this sooner given my interest in catastrophe bonds, but it is strange how seldom people in the prediction market ecosystem talk about catastrophe bonds. A prediction market contract with $1 million in volume is considered pretty large, but Artemis.bm tracked $6.7 billion in catastrophe bonds and insurance-linked security issuances in just Q1 of this year. There should be some transferable lessons, because every catastrophe bond implies a probability of an insured event: we'd expect a catastrophe bond for earthquakes in Minnesota to have a lower interest rate than one for hurricane damage in Florida. But I asked a couple of people at this conference about this and the impression I got is that catastrophe bond investors aren't doing serious forecasting on insured events. It should be possible to back out a Brier score of catastrophe odds derived from these bonds using historical data, but without digging into this, I'd expect there to be some finance research community breaking this all down.

2. Matt Parlmer said it better, but it was a surreal experience being at Manifest on Friday night when the Department of Commerce announced an export control directive banning foreign nationals from using Mythos 5 and Fable 5: some of the Anthropic employees at Manifest needed to react in real time to the order. Sam Roland practiced his 'DoW v. Anthropic: A Quasi-Inside Perspective' with me on Saturday morning to get some feedback, but in light of the news the organisers moved his presentation to the largest stage they had and there was still not enough room for everyone to sit.

3. David Shor's The Politics of AI talk was an alarming look at the state of public opinion. David Shor works at Blue Rose Research, which is a polling analytics and campaign research vendor affiliated with the Democratic Party; the firm runs a lot of polls and is known for its analytical rigour. While Blue Rose isn't a neutral actor, Shor gave a deep dive on the state of public opinion with very few normative arguments. According to the presentation, voters now support price controls by a 3:1 margin, which is up from only 20% in 2021. AI data centres are now less desirable neighbours than landfills and nuclear power plants, and only 29% of respondents would want to live next to one if it gave them free electricity. Shor rightfully pointed out that this was a pretty unreasonable perspective by the public. For a window of time between March and the Iran War, AI was the #1 issue for voters. While the public generally trusts business leaders more than Democratic Party electeds on economic issues, that wasn't the case when it came to AI: only 16% of those polled want no government intervention for AI-related job losses. Shor emphasised this is less popular than heroin legalisation. Lots of AI questions had similar responses across partisan affiliation, so the groundwork has been laid for bipartisan AI regulation.

4. Telecom is one of my special interests, and this weekend I made it a point to wear a Pacific Bell hat to send a 'please talk to me about telcos' message. One of the best conversations that I had this weekend was with a network engineer at a fibre ISP in the Bay Area; point-to-multipoint networks are quite different from the inter-data centre networks that I am more familiar with, and layer 2 is much more interesting when it means running fibre to people's homes. I'm leaving a lot out of this section because I need to do plenty of follow-up research; I took more notes on this conversation than I did for any official conference talk. Chance encounters like these are why I keep coming back to Manifest.

5. My hardest choice this weekend was between two talks in the same time slot: "Towards a Christian Machine Intelligence" from FAI's Tim Hwang and Bohan Lou's "How China's venture ecosystem differs from the West", and I ended up going to the latter. In the last couple of years, PRC regulators have made it more difficult for foreign venture capital to invest in semiconductors, robotics, and AI. With fewer venture dollars chasing the same companies, terms have gotten worse for founders: investments coming from regional governments incentivise short timelines to take companies public; otherwise, founders are on the hook to repay investments and may face criminal penalties. Even more founder-friendly private VC firms can now drive a harder bargain given the fundraising environment. Two other differences between the Chinese and American VC ecosystems are the attitude about investing in competitors and how introductions are made: Chinese firms will often back the most promising handful of companies in a vertical, while American firms generally don't invest in competitors. And because something like LinkedIn never took off in China, investors can't really be cold emailed, so warm introductions are the name of the game.

6. Jack Despain Zhou from the Center for Educational Progress had a thoroughly researched takedown of Jeannie Oakes' Keeping Track, an influential book that argues against grouping K12 students by ability in the classroom, such that all students are given the same curriculum at the same pace. Oakes grounds her entire argument in four pieces of education research. Three out of the four studies that Oakes cited either showed positive changes when students were grouped by ability, or didn't change curriculum and pacing across ability groups. The single piece of research that Oakes did interpret correctly was Marascuilo 1973, which showed that mixed-ability classrooms performed better on one of two tests given in the study after coaching the teachers of both groups. ² This is an incredibly flimsy justification, but literally thousands of school systems don't group by ability because of this book. Jack is a talented researcher and engaging speaker, and he's fighting the good fight for excellence in K12 education.

7. Most importantly, a sincere thank you to everyone who has helped make Manifest a reality. The event takes a ton of work, and I have only good things to say about a staff who works around the clock, outside vendors doing their jobs in a sometimes hectic environment, and volunteers taking time away from the conference to help make it all possible.