AIMS Lab

Samsung Labs has released LittleBit, a research method for compressing large language models below one bit per weight using latent factorization. The technique aims to shrink model memory requirements far beyond standard quantization, making large models cheaper to run on limited hardware. The release and code on GitHub are drawing attention from developers debating its practicality and how it compares to existing extreme-quantization approaches.

AMD has announced an $8.2 billion acquisition of World Labs, a move aimed at expanding into AI accelerators and model development. Commenters say the deal positions AMD to directly challenge Nvidia's dominance in AI chips and could intensify competition in custom silicon for AI workloads, shifting where enterprises train and run AI systems.

Humxn Research Labs says it will spend $50 million acquiring data on skilled trades, farming and manufacturing to train AI models for robotics. The company aims to source real-world task data from workers in these fields, an area where robotics training data remains scarce. The move highlights a growing push to pay for hands-on expertise rather than rely on internet-scraped content, drawing attention among robotics and AI watchers.

A free online textbook on AI Measurement Science from Stanford's AIMS Lab is drawing attention, ranking high on Hacker News. The resource focuses on how to rigorously measure and evaluate artificial intelligence systems, a growing concern as AI models are deployed widely and their reliability is debated. Readers are discussing the need for scientific standards in AI evaluation.

vCluster Labs has announced a managed cluster product built on NVIDIA's open-source Run:ai technology, according to a company release carried by business and tech outlets. The offering aims to simplify running virtual clusters for AI workloads by turning the open-source scheduler capabilities into a managed service. Coverage so far has been limited to the announcement itself, with little independent reaction or customer detail published yet.

A startup born from Penn State research is working to scale up domestic recovery of critical minerals in the United States. The venture aims to translate university lab work into commercial operations that recover minerals considered vital to supply chains for electronics, batteries and defense. The announcement, carried by EurekAlert! and Penn State, highlights growing US efforts to reduce reliance on foreign sources for critical materials.

The Wyss Institute at Harvard reports work on reversing biological ageing by reprogramming gene expression networks, an approach that aims to restore older cells to a more youthful state. Age-reversal research has drawn growing attention as scientists test whether partial cellular reprogramming can safely turn back age-related damage, with the Harvard institute highlighting this line of study in its recent communications.

Waters has announced that its Andrew+ robotic pipetting platform now supports native 12-channel pipetting, expanding the automation system's ability to handle higher-throughput laboratory workflows in fields such as genomics, drug discovery and sample preparation. The upgrade allows labs to run multi-well plate tasks faster with less manual intervention. Coverage so far is limited to trade press aimed at laboratory professionals.

The Global Innovation Lab for Climate Finance has endorsed eight new climate finance solutions that together aim to mobilize nearly USD 400 million for investment in emerging markets. The instruments, selected by the Climate Policy Initiative-hosted initiative, are designed to channel private capital into clean energy and climate projects in developing economies. The endorsements signal continued institutional effort to close the climate investment gap in countries where financing barriers remain highest.

A joking insult aimed at Proxmox, the open source virtualization platform, is making the rounds among Linux enthusiasts. The remark is affectionate rather than critical, the kind of teasing users direct at software they rely on heavily. Proxmox remains a popular free alternative for running virtual machines and containers on home labs and servers.

Lawrence Berkeley National Laboratory has developed an artificial intelligence model designed to dramatically accelerate simulations of earthquake ground motion. The tool aims to replace or augment computationally intensive physics-based modeling, allowing researchers to estimate how seismic waves will behave in a region far more quickly. The work could support faster seismic hazard assessment and improved earthquake preparedness for at-risk areas.

A Chinese company has released a large language model built specifically for legal work, according to legal tech publication Artificial Lawyer. The move adds to a growing wave of domain-specific AI models aimed at law firms and legal departments, competing with general-purpose systems from major AI labs. Details about the company, the model's capabilities and its target market remain limited in initial coverage.

Meta has introduced two new artificial intelligence models, Muse Wonder and Muse Wonder Full, describing them as its most capable systems to date. The models are aimed at agentic tasks, meaning they can act autonomously to complete multi-step goals rather than just answer questions. The announcement is drawing attention as Meta pushes further into the competitive agentic AI space.

Chinese AI company Zhipu AI has introduced GLM-Legal, a system aimed at advancing agentic legal intelligence, meaning AI agents that can perform multi-step legal tasks such as research and document analysis. The announcement, shared via the company's blog, is drawing attention among developers and AI watchers who are closely tracking the rapid progress of legal-focused and agentic AI tools.

Workday and Anthropic have announced a new partnership aimed at connecting AI reasoning to the business rules and processes that companies rely on, wherever employees work. The collaboration pairs Workday's enterprise software with Anthropic's AI technology, signalling a push to embed advanced AI assistants more deeply into everyday workplace and business operations.

French AI startup Mistral is preparing to release a new artificial intelligence model, according to The Wall Street Journal. The company, Europe's leading AI champion, aims to close the gap with dominant US rivals such as OpenAI, Anthropic and Google. The announcement fuels ongoing debate about whether Europe can build a serious competitor in frontier AI, with attention on Mistral's funding, compute access and ability to keep pace with American labs.

Applied Materials has opened a $5 billion research facility aimed at chipmakers across the industry. The lab, described by Fortune as one every chipmaker will need, is intended to help semiconductor companies develop and test next-generation manufacturing processes. The announcement is drawing attention as the chip industry races to build more advanced and efficient production capabilities.