Anil Madhavapeddy (@anil.recoil.org): I've had to respond to multiple OSS security issues recently and the wild thing is that agents can now generate exploits just on the *rumour* of a bug. This throws security embargoes out the window,…

A researcher trained a small transformer in 1.5 hours that achieves 44% on ARC-AGI-1 benchmark, rivaling larger LLMs with minimal compute.

Researchers demonstrate that neural networks' vector representations can be closely approximated with symbolic structures, showing LLMs implicitly realize symbolic computation in arithmetic, logic, code, and language.

Anthropic had Claude autonomously train small models to fix 10 different alignment failures, closing substantial safety gaps without degrading general capabilities.

Researchers introduce SCILAWS-BENCH, a benchmark with 118 problems across six disciplines to evaluate LLMs' ability to discover scientific laws from real data and synthetic hidden laws.

Detailed technical guide on diffusion language models, explaining their architecture, advantages over autoregressive models, and implementation techniques for text generation.