?? BrainWaves Newsletter 005

??Read time: 4 minutes

1. ?? Highlight of the Week

• ? Title: Simple Test-Time Scaling – from Li Fei Fei’s Team
• ?? Why it matters: This novel method enables significant performance improvements in LLMs with minimal cost and training time. By fine-tuning on a small dataset and applying efficient test-time adjustments, it achieves up to 27% gains on reasoning benchmarks, making advanced AI more accessible and cost-effective.

2. ?? Key Articles & Insights

• ?? 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. ??

Reference: Muennighoff, N., Yang, Z., Shi, W., Li, X. L., Fei-Fei, L., Hajishirzi, H., Zettlemoyer, L., Liang, P., Candès, E., & Hashimoto, T. (2025). s1: Simple test-time scaling. https://arxiv.org/abs/2501.19393v2

• ?? 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.

Reference: Sose, A. T., Joshi, S. Y., Kunche, L. K., Wang, F., & Deshmukh, S. A. (2023). A review of recent advances and applications of machine learning in tribology. Physical Chemistry Chemical Physics, 25(6), 4408–4443. https://doi.org/10.1039/D2CP03692D

3. ?? Tools & Resources

• ??Machine Learning in Production Course by Christian K?stner, 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/

4. ?? Social Spotlight

• ?? Spiked Systems for Colonic Drug Delivery: Architectural Opportunities and Quality Assurance of Selective Laser Sintering: As noticed in Orestis Katsamenis′ post, Selective laser sintering (SLS) 3D printing demonstrated its potential for advanced drug delivery by successfully fabricating spiked drug-loaded specimens that exhibited strong mucoadhesive properties and prolonged retention in the colon. This study highlights the capability of SLS to produce complex geometries without additional sintering agents, paving the way for personalized and more effective pharmaceutical treatments.

Reference: Angelos Gkaragkounis, Konstantina Chachlioutaki, Orestis L. Katsamenis, Fernando Alvarez-Borges, Savvas Koltsakidis, Ioannis Partheniadis, Nikolaos Bouropoulos, Ioannis S. Vizirianakis, Dimitrios Tzetzis, Ioannis Nikolakakis, Chris H. J. Verhoeven, Dimitrios G. Fatouros, and Kjeld J. C. van Bommel. ACS Biomaterials Science & Engineering Article ASAP. DOI: 10.1021/acsbiomaterials.4c02038

• ?? 300k+ Public Dataset Source: Finding datasets for ML projects is often a bottleneck, even as number of studies, technology and research capabilities continue to advance. Publicly available datasets are invaluable in such cases, and the Co-founder & CEO of Hugging Face recently announced that their platform now hosts over 300,000 public AI datasets spanning text, audio, image, video, 3D, time-series, tabular data, and more.

Source Link https://huggingface.co/

5. ?? Question for the Community

• ? What’s a concept or idea you once doubted but now strongly believe in?

6. ?? Upcoming Events/Opportunities

• ?? Artificial Intelligence (AI) Action Summit [10-11 February 2025]: The summit will held discussions on five different themes from Public Interest AI (how to develop infrastructure for social, economic and environmental outcomes) , Future of Work (how AI can help enhancing productivity and well-being), Innovation and Culture (how to build and deploy innovation ecosystems for cultural and creativity industries), Trust in AI (how to build AI trust using common scientific goals in security and safety), Global AI Governance (based on the existing studies such as UN, how to shape framework for AI governance).

Source Link https://www.elysee.fr/en/sommet-pour-l-action-sur-l-ia

• ??The Alan Turing Institute AI UK 2025 [17-18 Mar 2025] : Event will showcase cutting-edge AI and data science innovations, tackling challenges in defence, healthcare, sustainability, and beyond. The two-day event will feature 150+ expert speakers, 40+ interactive demos, and engaging sessions, including discussions on AI for decarbonisation, brain-computer interfaces, and the Isambard-AI research infrastructure.

Source Link https://www.turing.ac.uk/events/ai-uk-2025

?? Call-to-Action:

"If you enjoyed this, feel free to share it or reply with your favorite resource from the week!"