Right now the biggest barrier to AI-driven formal methods is that AIs are absolutely dogshit at coming up with good properties. I thought that was just a March 2026 thing but it seems to be a September 2026 thing too

Honestly it's still early fam. Enterprise stuff needs context and accountability that prompts alone can't give yet. Solo devs will get there though 🚀

I don't even know what to say, man

Starting to ask Claude to put together short videos to explain things I’m trying to learn.

To be clear, as we say in our latest Risk Report (xitter.cf/AnthropicAI/status/208…), I think the risk from present models is low. What I am worried about is superintelligence arising from recursive self-improvement, as we have said is happening faster than we thought (xitter.cf/AnthropicAI/status/206…).

Neural nets work.

Here is why I am afraid of AI. I understand how management, boards, investors think. They are all overextended, barely able to keep their head above the water. They've already outsourced their thinking to ChatGPT because they just can't keep up with the changing times. They do not understand why software development is so expensive. Their orgs are collapsing industry-wide as capital reallocates into the next thing, leaving virtually all companies exposed. Not just the VC backed, but cascading effects - layoffs are happening everywhere. Salaried managers are deeply motivated to continue to perform activities that generate their salary, and under this intense competitive pressure, it is all they can do "keep the music playing for one more quarter". And there is enough of a promise, from their perspective, a glimmer of hope, that AI will either help turn the ship around, or slow the fall, or even just fear that more agile competitors will use AI to defeat them. Even a clear-minded individual manager in an org must execute the plan handed down to them, regardless of their personal opinions -- they do not control the company strategy, and if they want to keep their tech salary, which is under attack from all directions, they bend the knee. 24% of US tech managers over age 50, once laid off, do not return to the workforce. Therefore, it doesn't matter what is real, it only matters what the majority believes, we are a democracy after all. The AI train has already left the station, it is accelerating, it is about to jump the tracks, and when it does, it will have far too high momentum to do anything about it, and we are all going to watch the slow-motion trainwreck take these managers out, and the entire tech sector with them, but a prerequisite to that happening is everything is going to break. The world is going to break. Climate change, financial crisis, inequality, cryptocurrency, energy, famine, war. And then the world will recover and be built anew, and the information age can finally begin. This is my 40 year prediction of what will happen in my remaining productive lifetime.