Brandon Gleklen

One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". At some point it's already pretty damn spherical and any improvement is marginal. Now of course smart humans aren't quite at the optimal bound yet on an individual level, and machines will have many advantages besides intelligence -- mostly the removal of biological bottlenecks: greater processing speed, unlimited working memory, unlimited memory with perfect recall... but these are mostly things humans can also access through externalized cognitive tools.

At this point, one way or another, super intelligence (at least domain by domain) seems like a given on any reasonable investment horizon. Might not happen, might be something we're missing, but it is hard to bet against it. So the real investment question is not about the production function for intelligence per se, it is about the demand function. Maxi case: Demand for intelligence is virtually infinite. The smarter the models get, the more we'll demand them. This applies to almost every conceivable domain, so the frontier model vendors will enjoy the same clear advantages they have in code as oligopolistic suppliers of a permanently capacity-constrained resource (frontier intelligence). Mankind will colonize planets, build mass solar arrays around the sun, apply intelligence with stunning ubiquity and completeness to manipulate our bodies, world, etc. Robotics will improve to the point where intelligence is embedded in the physical world, drawing massive compute. A true singularity + abundance scenario. Hard not to root for, except for the disruption on the way. Mini case: Demand for intelligence is capped in most domains. Humans evolved to have give or take the intelligence we "need" to operate on Earth/socially and we're limited in how much intelligence we can consume from external sources As a result, models will exceed the intelligence consumption capacity and/or requirement of the typical worker/consumer in 2-3 years, open source models will follow up shortly and the cost of intelligence will collapse to zero as the models get 100x more efficient in the years to come. Vendors with distribution (whether legacy or AI native) will seamlessly add intelligence to their products (the AI will help!). Much like electricity and the internet, intelligence will be something we take for granted as a feature of society, rarely think about and a commodity priced to input + distribution costs (energy, silicon, etc.). Of course, there are many, many cases in between- but the marginal question has clearly shifted from "is AGI/ASI possible" to "what are the implications once it is here." Current thinking is that this varies by domain- in some (customer support, certain enterprise agent applications) we're already arguably close to the intellectual horsepower required and we're seeing some companies start to migrate from frontier to frontier-y self-built models. In others (coding, strategy, hedge fund trading, etc.) we remain far, far away. In many others, we still lack the context to actually know exactly where we stand.

Current JDub OKC PG Pistons Blake ‘22 DeMar Rank ’em 1-4 🍿

Introducing Claude Managed Agents: everything you need to build and deploy agents at scale. It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days. Now in public beta on the Claude Platform.

This is really impressive! With the proper harnesses, and some guidance, we were actually able to find that same FreeBSD zero-day using Sonnet 4.6.