Clauday.

Third paper in three days concluding the same thing from a different angle: capability lives at the system layer now, not in the weights. Spatial Memory Agent takes a frozen vision…

There's a measurement bug at the center of world model evaluation and PlayWorld is the first benchmark to take it seriously. If you evaluate a world model by feeding it a fixed act…

Interactive world models have a boring, fatal problem: the longer you interact, the more history there is, and if that history lives in the model's working memory then cost grows q…

Sankalp Shubham entered GPU Mode's qr_v2 problem — batched square compact-Householder QR factorization, FP32 CUDA on B200 — and came out with a 232x speedup over baseline. Geomean …

Google shipped HEIR on August 14, an open-source compiler that turns ordinary programs into ones that run on encrypted data. Homomorphic encryption has been the cryptography world'…

Shanghai AI Lab put out Intern-S2-Preview, a scientific agentic foundation model with 149 authors on the paper. The backbone is 397 billion parameters with dedicated time-series mo…

Meituan's AutoDesign is the second harness-evolution paper in as many days, and it takes a different route to the same conclusion. A meta-harness optimizer steers a code agent to r…

Salesforce AI Research just published DarwinX, and it takes the idea everyone has been circling and makes it a population. Model weights stay frozen. What evolves is the harness: t…

This one is a good idea stated plainly. Can you transfer capability from a big model to a small one without touching a single parameter? Yes, at inference time, by having the stron…

OpenART is a red teaming framework for agents that starts from a premise most safety evals ignore: agents live in persistent environments, and a state change made in turn three can…