Matt's Thoughts In Between - Issue #2

Why are VCs so un-innovative?

Auren Hoffman has a great Quora post here, in which he points out that few venture capital firms follow the advice they give to their own portfolio companies. More broadly, VCs do very little to build long term equity value in their own firms. This is puzzling. Auren notes that the top private equity firms do do this, which seems like a hint that it's not impossible.

Traditionally VC has been a business that requires superstar partners to attract and pick the best investments. And superstars, understandably, want to capture most of the upside from what they do - which doesn't leave firms much room to invest in the future. That's why firms that are exploring ways to do VC systematically are so interesting. Social Capital's recent "capital as a service" experiment is fascinating - and Y Combinator and Angel List are arguably two of the most important disruptions to VC over the last ten years. And, of course, that's exactly what we're trying to do at EF and why we raised venture capital ourselves.

Computational power and AI nationalism

Jack Clark's must-read ImportAI newsletter flags an important development this week: that computational power - and not data - is now the most important driver of AI research. This is something I'm hearing more and more from smart people in the field. It's strongly linked to a idea that's also becoming more prominent: "AI nationalism" or AI as a national security issue. Nation states ought to be able to build big advantages in computational power compared to almost all non-state actors, which opens the door to a new kind of geopolitical competition. Axios has a good short summary of some of the key issues. I'm sure it's a topic we'll be revisiting.

The persistent importance of place

There's a lot of talk about the importance of place in politics at the moment (e.g. the idea of "Somewheres" and "Anywheres" in Brexit). What's astonishing is how long place persists, despite large-scale migration over the last few centuries. I read last week that the presence of Protestant Huguenots in the 16th century is still a strong predictor of a French region's Socialist vote today (Robert Tombs shows something similar about Anglicanism and the UK Conservative Party in his excellent History of the English).

But then I was listening to this excellent podcast with the anthropologist Joseph Henrich and it turns out that a few hundred years is small fry. This research came up in the conversation - and it suggests that economic development even in 1000 BC is a strong predictor of economic development in the same place today. Intuitively, it's hard to imagine this persisting another few centuries... but then intuitively it shouldn't have survived the last few either.

Quick links

The ubiquitous Canon in D: I loved this podcast (from a year ago) with Derek Thompson, author of Hit Makers, which in turn led me to this hilarious / surprising video of 40+ pop songs based on the same four chords
Enlightenment Redux: Steven Pinker's new book is prompting a lot of discussion of the Enlightenment; this Twitter thread is one of the best (and apparently Pinker is not into AI safety)
Google Maps never dies: Fascinating exploration of the "insane backwards compatibility of Google Maps"
How to win a lot of bets: Bryan Caplan reflects on why his long-run predictions have so often been right
Why bots are everywhere: Interesting tweet that makes a good argument that Twitter is actually really good at removing malicious bots

Your feedback

Thanks for all your comments last week. Feedback always welcome. If you like this, please tell a friend or share the subscribe link: https://www.matthewclifford.com/aboutme/

Until next week,

Matt