The AI Reader.
Tuesday, August 4, 202647 min read
Working Knowledge · The Instruments

GPT-4 sits above the EU's systemic-risk line on a number OpenAI never disclosed

Training compute is the FLOP count regulators now write into law to mark the AI frontier, and almost no lab measures it: the headline figure is reconstructed after the run, two ways, each an estimate with a loose joint.

10 min read12 sources
Working Knowledge · What Could Go Wrong

AI's strongest bug-finding results so far came out of defensive research

Google's Big Sleep found a real flaw in SQLite and DARPA's finalists found eighteen more, every one surfaced by a team whose job was to fix it.

10 min read9 sources
Working Knowledge · The Evidence

AlexNet won ImageNet 2012 on two gaming GPUs

Krizhevsky, Sutskever, and Hinton beat the runner-up by roughly ten points in 2012, a margin that pushed computer vision toward depth, data, and compute for good.

9 min read8 sources
Working Knowledge · The Mechanics

A chatbot quotes your PDF without learning a word of it

To answer, the model rewrites your file as points in space and reads back the few passages that land nearest your question.

9 min read9 sources
Working Knowledge · When AI Breaks

nH Predict sets one patient's discharge date from six million others

A class action against UnitedHealth says an algorithm's length-of-stay estimate became the deadline that ended nursing coverage for elderly patients whose doctors wanted them to stay.

9 min read9 sources