Lots of details about the enshitification of Quora. Focus on SEO, loss of moderators, AI-generated answers, and overwhelming advertising have utterly killed it.

There are a surprisingly large number of cases of people claiming an AI did something when it’s really done by a person. Sometimes it’s a lie, sometimes they fooled themselves.

AI is now (nearly) capable of being used to judge gymnastics.

An old version of ChatGPT will crash if you include certain words in the input.

This AI, nothing more than a Stanford student project, can look at the background of a photograph and guess the location with quite good accuracy.

Terry Tao gives a detailed description of exactly how to use Lean to formalize a proof.

We’re regulating AI wrong because we understand the risk wrong. It’s not the paperclip maximizer we need to worry about, it’s a corporation using an AI you will trust but using it to achieve the corporation’s goals. Fiduciary requirements and open models are the solution.