Manifest 2024 & Squiggle
%20(2).png)
Manifest 2024 was this last week, along with Manifest Summer Camp. You can see the schedules for these here.
I gave two Squiggle workshops, one Squiggle presentation, and three focused events. Each session was around 60 minutes.
I thought that the Squiggle workshops went well, though they were introductory. I've gotten used to these. The short version is an overview, with a lot of back and forth on specific audience questions. These events change significantly based on the audience in question.
The Manifest audience seemed to understand the basics fairly quickly. My main push for this event was to get feedback and encourage scorable function forecasting. I find that the basic concept here can be quite tricky for some people to grasp, but I think a lot of this audience (at least, the people who spoke up or asked questions afterward) understood it. My hunch is that if scorable function competitions are run, many people in this crowd wouldn't be too intimidated to join. This is the first big hurdle!
Related, I also spoke about Squiggle and relative value functions, and that also seemed to make sense to the audience. Note that I did present on these topics at my talk last year at Manifest, albeit in lesser detail, but I get the impression that only a small slice of my target audience watched or remembered that talk. Eventually, I hope that our epistemic infrastructure will enable people to say ideas once and have them translate to everyone who would find them useful, but until then, ideators have a whole lot of repeated communication to do.
Ideally, organizations at Manifest could eventually run their own scorable function forecasting tournaments, but the bar to organize a competition is much higher than that for forecasting directly. Most attendees were either forecasters or people in charge of commercial forecasting platforms - there weren't many experimental organizations who might run small-scale experiments soon.
My guess is that very few of the big prediction market platforms will use scorable functions anytime soon, but it is possible that one or two prediction tournament platforms might. At the very least, it seems like a clean improvement to prediction AI systems. I chatted with a few organizations that run tournament platforms and AI platforms; hopefully, there will be positive outcomes from those discussions in the upcoming months (fingers crossed!).
My "focused events" were on:
1. "Crafting great AI strategy forecasting questions, with scorable functions"
2. "Ambitious Epistemic Technologies"
3. "Estimate Key AI Variables, with Squiggle"
These events had 8-20 attendees each. For the first two, we had people comment and discuss questions in Google Docs, for the last one, we had people split up into teams and make forecasts using Squiggle.
In the "Ambitious Epistemic Technologies" talk, members said that they were unfamiliar with the terms "Epistemic Technologies," "Epistemic Infrastructure," or similar, but they seemed to understand them quickly once explained. I imagine there's a lot more promotion and thinking to do here.
My primary formal presentation was a comprehensive overview targeting a technical audience. The notes for this session can be found here, and I've been informed that videos should be available within the next 2-4 weeks. In retrospect, I may have been overly ambitious in trying to cover a wide range of material, which resulted in a rushed pace and me straining my voice in the process.
In the future, it would be interesting to explore the possibility of conducting more extensive 3-10 hour workshop sessions, if we could find the right space and people for it. This extended format would allow for a thorough exploration of the main ideas while providing participants with ample opportunities to directly try out the ideas for their own interests.
As to my general thoughts on the conference, I've written about them here. I very much appreciate the team that put this on.