A newsletter breaking down AI research, technology, and Australian property in plain English
XGBoost security classifiers appear near-invincible against gradient attacks but crumble under score-based methods, while SHAP explanations break even when pred
DynaKRAG treats multi-hop evidence gathering as a learned control problem, outperforming fixed RAG pipelines on three benchmarks with Qwen2.5-7B-Instruct.
A new arXiv preprint proposes letting each user in a decentralized network set their own privacy budget for graph data, fixing a uniform-noise flaw that degrade
Hugging Face's new kernel repository type brings native compute code to the Hub with trusted publishers and code signing — changing whose code your machine runs
A new preprint shows matrix-structured optimizers train atom-level AI models faster and more accurately than Adam, especially when labels are scarce.
A 22-language study finds prompting LLMs to reason in English while answering in other languages substantially closes the uncertainty gap for low-resource tongu
Budget-aware RoR policy splits per-query inference spend between resampling and rerouting across 11 open-weight LLMs, beating five baseline strategies.
New arXiv preprint PeTeR proposes a data-free way to harden pre-trained probabilistic circuits against distribution shifts without retraining from scratch.
A new arXiv preprint uses agentic topology sampling to predict building IoT sensor readings with zero building-specific training, matching fully trained models
