OpenAI Just Released a Model Too Dangerous For You To Use
7 frameworks to make sense of this moment.
OpenAI released a new model this afternoon that is testing nearly as powerful yet cheaper to use than the Mythos class models. The federal government considers both this model and Mythos too dangerous for public use and is, for now, restricting access to a list of “government-approved partners.” Outside of specific defense-adjacent technologies, I can’t recall a Republican administration pivoting this hard into central planning. While this matters in terms of politics, that is not what this newsletter is meant to cover. No, it matters for you because of a very specific word—it is unprecedented. This is something wholly new.
In my writing on the Mythos class of models, I’ve been arguing that here is where the “training wheels come off.” Meaning that we have reached the point in the improvement curve where these models progress from pretty useful technology to something that can agentically do work, hack into classified government systems, and alter the nature of information on the internet. Let me be very blunt: this shit is happening. It is real. It is only going to go faster and crazier from here. In that light, I understand the knee-jerk reaction of the administration to shut everything down until they can figure out what to do. I wish them the best of luck in this unprecedented challenge.
There it is again; that word bubbled out of me organically. (I admit this with some degree of writerly shame). Despite being a little overwrought, it is the right word for this moment because our future is one of dramatic change such as we haven’t experienced in 100 years. It may be another industrial revolution, or if we reach AGI, it may be something even wilder. But in either scenario, you should prepare for the plan you made for your life, career, company, whatever, to be torn asunder.
I’ve spent the last four years of my life, burning tens of thousands of hours to write millions of words, all trying to grapple with that change. The result of that effort is 7 original frameworks and mental models that have led readers to describe me as spooky right.
I don’t know your life. It is hard for me to prescriptively tell people what to do with this new world of AI capability because no advice is universally true. Instead, let me give you the tools that have helped me make sense of all this. Charlie Munger spent his life insisting that wisdom is not a pile of facts but a structure he called a latticework: “You’ve got to hang experience on a latticework of models in your head,” he said in 1994. These 7 models are the ones that have helped me make sense of all this.
