Claude 3.7 Sonnet's 59-minute time horizon is where its success crosses one half
METR built the number from 170 timed software tasks, and its seven-month doubling, drawn past the last released model, became a public calendar for automating a month of work.
Give GPT-4 the same options in a new order and it changes its answer
The swing runs from 13 to 75 points across five benchmarks, and the two papers that measured it most carefully split on whether position or the answer letters drive it.
Pricing algorithms learn to hold prices high, so far only in simulation
Two Q-learning agents reached about 85% of the way to monopoly pricing with no instruction to coordinate, and the economists who reopened their code doubt it would survive a real market.
Regulators kept ruling Clearview illegal as its face database grew past 30 billion photos
Clearview turned billions of public photos into biometric codes and sold face searches to police, and the European fines against it have gone largely uncollected.
Google's 2017 layer split a model's size from what it computes
Noam Shazeer's team sent each token to a few expert sub-networks out of thousands, the design behind why Mixtral today lists 47 billion parameters but runs only 13 billion per token.