Everything the labs are doing points to an economy sized data bottleneck. Literally. The learning rate of economically useful AI is bottlenecked by the learning rate of the economy. This is endogenous, but it is bottlenecked by effective deployment. In other words, diffusion isn't cope, it's the whole thing! AI can only learn as fast as its ability to ascertain state, judge (a priori) the highest ROI intervention, perturb state, and realize whether the perturbation generated a positive (adaptive) or negative outcome. This, in many areas (and unlike code and math) will entail organic time. This is also why you see more math and software engineering prodigies than in other fields. Getting really good at math doesn’t require access to permissioned local information environments and having the right to actuate within those environments to ascertain long horizon feedback.

I will be returning to @MTSlive at 10am PT today with @ByrneHobart to discuss our @arenamag piece on Palantir, the Forward Deployed Model, institutional fitness in an age of accelerating machine intelligence, the bottleneck in converting tokens into economic value, and lab positioning in this world. Looking forward to it!

Progress in technology is all about breaking false tradeoffs. Palantir, has been forged, from the beginning, to help critical institutions leverage the accelerating power of digital computation without sacrificing their sovereignty or the sovereignty of their constituents. This was prescient and is more important than ever. As, today, some in Silicon Valley believe "commoditizing your complement" means commoditizing institutions and humanity itself. In some sense, the effective accelerationists of Silicon Valley believe technology is God and the effective altruists believe they are God. The former believe that sufficiently powerful technology will automatically organize society in the way that threads the needle between the Antichrist and Armageddon, and the latter believe they are the only ones capable of creating technology and organizing society in a way that threads that needle—convienently, according to their own utility function! The AI successionists don’t even care whether we thread that needle at all. Palantir gives us a different option: technology as an unapologetic accelerant of human curiosity and agency, yes, but also of institutional fitness. In this view, we don't have to surrender our economic sovereignty to an all-knowing model, nor cede our moral authority to the unelected engineers in SF and the well-meaning, but generally histrionic “philosophers” in Berkeley who built it. Instead, we can use ever-more powerful computation to make the pro-human institutions we select (through free markets and the democratic process), and therefore our society, radically more capable and resilient.