Google Launches Gemma 4, a Powerful Open-Weight AI Model
Google has unveiled Gemma 4, its latest open-weight AI model featuring powerful multimodal capabilities that support audio, text, and image inputs. Building on the success of its predecessor, Gemma 3, this new model is engineered with advanced reasoning to ’think’ before responding, making it a potentially vital tool for developers. A key feature of Gemma 4 is its efficiency, allowing it to run on smaller machines and increasing its accessibility across the development community. This release signifies a major step in the trend toward creating more versatile and efficient generative AI models.
Sources:
Researchers Use Uncertainty to Improve AI-Based Molecular Design
Scientists from the U.S. Department of Energy’s Brookhaven National Laboratory and Texas A&M University are pioneering a new method to enhance AI-based molecular design by incorporating uncertainty. This innovative approach enables AI models to generate molecules with superior predicted properties compared to their predecessors. The research, featured on the cover of Molecular Systems Design & Engineering, utilizes Variational Autoencoders (VAEs) to intelligently explore and identify promising molecular structures. By embracing uncertainty, these AI tools can more effectively design molecules for critical applications in new drugs and advanced materials, offering a way to fine-tune generative AI models without costly and time-consuming retraining.
Sources:
Waymo and Waze Team Up to Detect Potholes in US Cities
Alphabet subsidiaries Waymo and Waze have launched a pilot program to identify and report potholes across major U.S. cities, including San Francisco, Phoenix, and Los Angeles. The initiative leverages data collected by Waymo’s autonomous vehicle fleet to pinpoint road imperfections. This information is then relayed to city officials to facilitate repairs and shared with Waze users to help them avoid road hazards. This collaboration is a significant example of using autonomous vehicle data and smart city technology to improve urban infrastructure. To date, the program has successfully documented approximately 500 potholes.
Sources:
Generative AI Sees a Divide Between Open Source and Proprietary Models
The generative AI landscape is showing a clear split between open-source and proprietary model release strategies, exemplified by recent moves from Zhipu AI and Anthropic. On April 7, 2026, Zhipu AI released GLM-5.1, a massive 744-billion-parameter mixture-of-experts model, under the permissive MIT license. On the same day, Anthropic confirmed its most powerful model, Claude Mythos, but stated it will not be publicly available. Instead, access is restricted to fifty organizations via “Project Glasswing” for defensive infrastructure vulnerability scanning. This divergence highlights a growing philosophical debate over access to powerful AI. The first week of April 2026 was busy for AI, with eight model releases, including Google’s Gemma 4 family and Alibaba’s Qwen 3.6-Plus.
Sources:
JPMorgan Downgrades Nutanix, Citing Cooling Cloud Infrastructure Market
JPMorgan has downgraded Nutanix stock from Overweight to Neutral, setting a new price target of $44. The downgrade is based on the assessment that the rally in the cloud infrastructure market may be slowing for software-centric companies. Analysts cite a volatile macroeconomic outlook for late 2026 and 2027, along with rising competition from other hybrid cloud vendors, as factors limiting Nutanix’s near-term growth. Despite a recent earnings beat, the company’s stock has experienced a significant year-to-date decline, indicating that the market had already begun to anticipate this valuation adjustment.
CNCF and Kusari Collaborate to Enhance Cloud-Native Supply Chain Security
The Cloud Native Computing Foundation (CNCF) has partnered with Kusari to strengthen software supply chain security for its portfolio of cloud-native projects. This collaboration grants CNCF projects free access to Kusari Inspector, an AI-powered tool designed for code review and dependency management. The initiative aims to give project maintainers greater visibility into their software supply chains, enabling them to identify and mitigate risks within complex dependency ecosystems. This partnership directly addresses the escalating security challenges in modern software development, such as AI-generated code and sophisticated supply chain attacks, by integrating security tooling into developer workflows to build a more resilient open-source ecosystem.
Sources:
Pasqal and True Nexus to Revolutionize Food Science with Quantum Computing
Quantum computing firm Pasqal is collaborating with computational intelligence company True Nexus to apply neutral-atom quantum processors to challenges in the food industry. The partnership will focus on developing a detailed, dynamic 3D model of protein gelation, a key process influencing food texture. This quantum-powered modeling is expected to overcome the limitations of classical computers in understanding complex protein functionality. By integrating data on molecular structure and processing conditions, the project aims to transition the alternative protein sector from a trial-and-error methodology to a more precise, design-driven process. The ultimate goal is to create a reference model to accelerate the development of new sustainable protein sources. This news follows Pasqal’s announcement of plans to go public via a merger with Bleichroeder Acquisition Corp. II.
Sources:
Q-Factor Emerges from Stealth with $24M to Build Million-Qubit Quantum Computer
Israeli startup Q-Factor has emerged from stealth with $24 million in seed funding to develop a quantum computer with over one million qubits. The funding round was co-led by NFX and TPY Capital, with participation from Intel Capital. Founded by four physicists from the Weizmann Institute of Science and the Technion, Q-Factor aims to commercialize decades of atomic physics research. The company is utilizing a neutral atom-based approach, which allows quantum information to be stored for longer durations and controlled with light, eliminating the need for extreme cooling or complex wiring. Q-Factor’s architecture is designed for scalability, aiming to overcome the size limitations that prevent current quantum computers from tackling most commercial applications.
Sources:
PyTorch Foundation Expands AI Stack with Safetensors and Helion
The PyTorch Foundation, part of the Linux Foundation, is expanding its open-source AI stack with the addition of several key projects. Safetensors, a tensor serialization format from Hugging Face, joins to mitigate security risks by preventing arbitrary code execution in model files—a known vulnerability in older formats. Helion, a project donated by Meta, is being integrated to simplify and standardize the development of custom machine learning kernels. Additionally, ExecuTorch is being incorporated into PyTorch Core to enable efficient AI inference on edge and mobile devices. These additions are designed to provide AI developers with a more secure, interoperable, and comprehensive toolset for the entire machine learning lifecycle.