Our Most Controversial Thesis: Open Source SuperIntelligence Will Be Widely Available
How to invest when everyone has ASI
Happy Sunday and welcome in Investing in AI. Be sure to check out our AI Innovator’s Podcast for the latest interviews with AI pioneers. And if you happen to be NYC based, you can apply to attend our private AI in PE dinner coming up May 22nd, where we will have a panel of experts who have actually bought companies and applied AI transformations to them. We have limited space and are almost filled up, but if you are an Investing AI reader I’ll give you priority. (We also have room for one more dinner sponsor).
Today I want to discuss AGI, or maybe ASI. When you talk to most people about Superintelligence, there is this idea that its a race because whoever gets their first is at a significant advantage. The common thought process behind this idea is that, a superintelligence can start improving itself faster and faster and, no one else can catch up. I’ve written before about why I think this idea is wrong, but today I want to go a step farther.
We have an active investing thesis now at HalfCourt Ventures that says this - whenever Superintelligence is invented, within a month other labs will have it and within 4-6 months it will be open sourced and widely available. This is not a common view in tech circles but, we hold it for a few reasons.
One is, open source AI is getting better and better. The time from closed source models launching to open source matching it runs anywhere from 5-22 months, but is getting shorter and shorter. The second is, there will always be compute and other physical constraints on these models that will matter just as much, or more, than the algorithm and data.
As with many early stage tech bets, perhaps we are wrong. But what if we are right? What if, 6 months after its invention, everyone everywhere has access to superintelligence and the ASI itself is not an advantage? Where do you want to be invested, and what assets to do you want to have?
That is the main question we have been asking ourselves.
We are still working through it but I have a few partial answers.
You want to own the best tools for monitoring, analyzing, testing, and containing a superintelligence.
You want to be the most effective and efficient at running superintelligence on a given AI system or architecture.
You want to own workflows that have been difficult to automate but can be automated by a true superintelligence.
My job here is to make you think a bit and not to spoon feed you investment ideas, so I’m not going to give any specifics yet about the areas we are targeting for these points. Plus its early and we need more time to think through this.
But with the world focused on the value of creating AGI/ASI, I want to ask you as an investor in AI, is it possible you make more money in the companies and sectors around this rather than the AGI itself? Why do we believe one company will control this? Invert the idea and assume its widely available. How will you invest then?
I know some of you will hate this idea so, if you hate it enough to submit a counterpoint piece that is well thought out, I’ll publish it here. Thanks for reading.
I think there's strong logic to your position. Every day we're getting more data about open source catching up. Very compelling. And it gives me a sense of relief.
I appreciate this outline of your thesis. The lack of standard definition for what constitutes ASI/AGI makes it challenging to write about. In a world of true superintelligence I'm unsure how any human investors can (legally) keep finding alpha.