AI News
RBS-Attention Claims a 20x Prefill Speedup Without Retraining a Single Model Weight
September 21, 2026
A new training-free sparse-prefill method claims up to a 20x attention speedup for long-context LLMs by fixing a subtle flaw called mean dilution, and it barely dents accuracy in the process.
Read more
NVIDIA Retires GenAI-Perf, Ships AIPerf to Fix a Benchmarking Tool That Was Lying to Engineers
September 20, 2026
NVIDIA has replaced GenAI-Perf with AIPerf, a multiprocess benchmarking tool built to stop the load generator itself from becoming the bottleneck it’s supposed to be measuring around.
Read more
Stable AI’s LimiX-2 Crams Three ML Jobs Into One 400M-Parameter Model
September 19, 2026
Stable AI’s LimiX-2 packs classification, regression, and missing-value imputation into one 400M-parameter model, trained on synthetic causal data and released open-source after beating AutoGluon 1.6 and TabPFN-3 on three major benchmarks.
Read moreAnalysis
Guides
Cross-Post 5 Platforms in 15 Minutes: The Typefully Workflow
May 22, 2026
Native posting across 5 platforms = 60 minutes. Typefully cross-posting = 15. The 6-step workflow that saves 20 hours/month, and the 4 common mistakes to skip.
