Curated insights and tools for curious minds.
📅 Date: 09/02/2025
📰 s1: Simple test-time scaling – Li Fei-Fei and her team introduced a novel LLM test-time scaling method that can be trained in less than 30 minutes for $50. By fine-tuning Alibaba Cloud’s Qwen on a carefully selected 1K-row reasoning dataset and using budget-friendly test-time tweaks—like early termination or adding “wait” to adjust thinking time—they claim up to 27% gains over OpenAI’s o1-preview on MATH and AIME24 benchmarks. 🚀
Source Link https://arxiv.org/abs/2501.19393
📖 A review of recent advances and applications of machine learning in tribology – This article reviews ML studies comprehensively and highlights how they are helping to analyze vast amounts of underutilized experimental and computational data to uncover complex structure–property relationships and optimize lubricant design efficiently. By leveraging neural networks, supervised learning, and stochastic approaches, researchers can model non-linear tribological behaviors, improve material performance, and accelerate discoveries in friction, wear, and lubrication studies.
Source Link https://pubs.rsc.org/en/content/articlelanding/2023/cp/d2cp03692d
🔧Machine Learning in Production Course by Christian Kästner, Carnegie Mellon University: Spotted via @Alejandro Saucedo’s post—this free course covers everything you need to deploy ML models into production. Clear, structured content guides you through ensuring quality, scaling, and successfully maintaining models. Worth checking out!
Source Link https://mlip-cmu.github.io/s2025/
🛠️ Psychic LaTeX Generation Tool: I often need to convert my keyboard-typed math formulas, which may contain human errors, into flawless, well-formatted LaTeX expressions, and this free tool makes that process effortless.
Source Link https://psychic-latex.vercel.app/
What the authors aimed to achieve.