Research
Papers, benchmarks and the techniques behind what eventually ships.
-
Anthropic used Claude to speed up 30 open biology models
Anthropic reports roughly 4x average speed-ups after Claude spent four weeks optimising more than 30 open-source biology models. All figures are its…
-
Anthropic puts a number on how much AI research Claude leads
Anthropic says Claude now leads 26% of its own AI research work, up from under 1% in February, and wants rival labs…
-
Google open-sources a way to fake the freight data nobody will share
MilleMiglia generates synthetic freight networks so researchers can benchmark middle-mile logistics without the proprietary data carriers keep private.
-
A proof about what shallow transformers cannot compute
An IBM Research team has proved that constant-depth transformers and diffusion language models cannot match shallow quantum circuits on two specific tasks.
-
OpenAI sets out how it will disclose model misalignment, and publishes six cases
OpenAI has published a standing process for investigating and disclosing misaligned model behaviour, together with six cases observed over the past six…
-
An AI agent scoring 77% on average solves only 53% of tasks every time
IBM Research puts a figure on agent flakiness: an agent scoring 77.4% averaged across five runs completed only 53.0% of tasks on…
-
Dropout curbs overfitting
Hinton and colleagues publish dropout, randomly disabling units during training.
-
Deep networks take over speech recognition
Four labs jointly report deep neural networks beating existing speech recognisers.
-
Hinton finds a way to train deep networks
Hinton, Osindero and Teh publish layer-wise pretraining for deep belief networks.
-
Breiman formalises the random forest
Leo Breiman publishes Random Forests, combining many decision trees into one predictor.