Long-form arguments that took more than one sitting to make.
Each one starts from a question I could not answer quickly, sets a condition under which the argument would fail, and then runs it against the record. Where it lost, the piece says so and shows the working. The shorter, faster writing lives on the blog. This is the shelf for the ones that needed the room.
Where an essay makes a prediction specific enough to be wrong, it gets scored in public: the Reformation Watch checks its seven indicators daily.
Latest
Andrej Karpathy says the AI gap is now a sharp funnel. Here is what the middle of it looks like from a small business, and how to cross it without betting the company.
A falsifiable prediction stops being falsifiable the day its author keeps the ledger. So the watcher publishes on its own now, and the first thing it published is about who keeps Anthropic's.
Every durable institution is a prosthesis for a virtue the population does not have. Which means the demand to become mature enough to agree, before we build anything, has the causality backwards.
Effective altruism had the shared text, the lifelong pledge and the costly signal the sociology of sects predicts. It also had the board seats and the cheque book. It lost both, one November apart.