Daily-Digest

Anthropic Launches Bloom: An Open-Source Framework for AI Model Evaluation Anthropic has launched Bloom, an open-source agentic framework engineered to automate behavioral evaluations of advanced AI models. This innovative tool enables researchers to define a specific behavior and then automatically generate numerous test scenarios to measure its frequency and severity. The primary goal of Bloom is to accelerate the creation of high-quality safety and alignment evaluations, which are often labor-intensive and can quickly become obsolete as AI models advance. Anthropic has already utilized Bloom to build and release benchmark results for four key alignment-relevant behaviors across 16 different frontier models. The framework also integrates with popular tools like Weights and Biases for large-scale experiments and complements Anthropic’s other open-source evaluation tool, Petri. ...

Google DeepMind to Launch AI-Powered Automated Science Lab in UK for Superconductor Research Google DeepMind will establish its first automated research laboratory in the UK in 2026, expanding its partnership with the UK government. The lab will utilize AI and robotics to accelerate scientific discovery, with an initial focus on developing new superconductor materials. This advanced facility will be fully integrated with Google’s Gemini AI. The partnership aims to reinforce the UK’s position as a global leader in science and technology. As part of the collaboration, UK scientists will receive priority access to some of DeepMind’s cutting-edge AI models, including AlphaGenome for DNA sequencing. The initiative will also explore using AI for cleaner fusion energy, developing safer AI educational tools, and enhancing public service efficiency. ...

New AI System Enhances At-Home Balance Training for Physical Therapy Researchers at the University of Michigan have developed a new machine learning model to support home-based balance training for physical therapy patients. The system utilizes data from four wearable sensors to predict how a physical therapist would assess a patient’s performance on balance exercises. This AI-driven feedback can help patients make faster progress and maintain their abilities after their formal therapy sessions have concluded. The model was developed by training convolutional neural networks with sensor data and evaluations from physical therapists, achieving nearly 90% accuracy in predicting patient balance ratings. The technology aims to make balance training more accessible, particularly for those in rural areas, and to assist physical therapists in making healthcare decisions. ...

MIT Unveils BoltzGen: A Generative AI for Designing Therapeutic Molecules In a major leap for drug discovery, scientists at MIT have introduced BoltzGen, a generative AI model capable of creating novel protein binders from scratch. This innovative tool is designed to address diseases that have long been considered difficult or impossible to treat, marking a significant shift from predictive biology to generative engineering in medicine. How BoltzGen Unifies Design and Prediction BoltzGen stands out by unifying protein design and structure prediction into a single, powerful process. The model incorporates specific constraints to ensure that the molecules it generates are not only novel but also functional and stable, adhering to the fundamental laws of physics and chemistry. Unlike previous models that were often limited to specific protein types, BoltzGen is engineered as a more general-purpose tool, capable of creating a much wider range of therapeutic molecules. ...

OpenAI Releases GPT-OSS, Its First Open-Source Large Language Models In a significant move for the AI and open-source software communities, OpenAI has released GPT-OSS, its first-ever family of open-source large language models. Licensed under the permissive Apache 2.0, this release introduces two powerful models: gpt-oss-120b and gpt-oss-20b. Both are engineered for efficient inference and strong reasoning, leveraging a Mixture-of-Experts (MoE) architecture. The larger gpt-oss-120b model rivals OpenAI’s proprietary o4-mini and operates on a single 80GB GPU. Meanwhile, the compact gpt-oss-20b model, comparable to o3-mini, is optimized for on-device AI applications, requiring as little as 16GB of memory. Complementing the models, OpenAI also launched ‘Harmony,’ an open-source project defining a new prompt template format. ...

Tech News Digest for July 30, 2025 A roundup of today’s top stories, featuring Anthropic’s massive new funding round, the emergence of powerful open-source AI from China, and significant progress in the fields of quantum computing and autonomous systems. Artificial Intelligence & Machine Learning Anthropic Reportedly Raising $5B at a Staggering $170B Valuation Generative AI company Anthropic, known for its AI assistant Claude, is reportedly in the process of closing a $5 billion funding round that would value the company at $170 billion. This new valuation would place Anthropic ahead of Elon Musk’s xAI, though still behind OpenAI. The round is said to be led by Iconiq Capital, with potential participation from the Qatar Investment Authority and Singapore’s GIC. This funding surge highlights the intense investor interest in the generative AI space and the escalating valuations of its key players. ...

Xanadu and HyperLight Announce Photonic Chip Breakthrough for Quantum Computing Xanadu and HyperLight have achieved a significant milestone in the development of thin-film lithium niobate (TFLN) photonic chips, which are critical for scaling photonic quantum computers. Through a joint effort, the companies have demonstrated waveguide losses below 2 dB/m, a new industry benchmark for performance. This advancement results in an electro-optic switch loss of approximately 20 milli-decibels (mdB), one of the lowest values ever reported for this type of component. The breakthrough is crucial for building future generations of fault-tolerant, utility-scale photonic quantum computers. This achievement builds on the technology used in Xanadu’s “Aurora,” the world’s first fiber-networked photonic quantum computer. ...

Kubernetes Introduces Post-Quantum Cryptography Support In a significant move to future-proof cloud-native security, the Kubernetes project has announced support for post-quantum cryptography (PQC). This enhancement integrates a hybrid key exchange mechanism through the Key Management Service (KMS) plugin system, marking a critical first step towards crypto-agility in the face of emerging threats from quantum computers. Quantum computers pose a long-term risk to current encryption standards like RSA and ECC, which are fundamental to securing TLS and X.509-based authentication within Kubernetes. By adopting hybrid key exchange algorithms, Kubernetes can now gradually transition to post-quantum secure schemes, ensuring long-term data protection against “harvest now, decrypt later” attacks without disrupting existing cluster operations. This development aligns with a broader industry trend, as major tech companies like Meta, Google, and Cloudflare are also actively integrating quantum-resistant protocols. ...

Tencent Unveils Open-Source 3D World Generation AI Model At the World Artificial Intelligence Conference (WAIC) in Shanghai, Tencent launched its Hunyuan 3D World Model 1.0. This open-source AI model can generate detailed, interactive 360-degree 3D environments from simple text prompts or single images. The model is designed to be fully compatible with standard 3D graphics and animation workflows, known as “CG pipelines,” simplifying the creation of virtual reality experiences and video games. By making the Hunyuan3D model accessible to global developers, Tencent aims to democratize advanced content creation tools and foster open collaboration. This release positions Tencent as a major competitor in the race to develop foundational AI for immersive content. ...

NVIDIA Releases Llama Nemotron Super v1.5 for Advanced AI Reasoning NVIDIA has launched the Llama Nemotron Super v1.5, a new open-source model that sets a higher standard for reasoning and agentic AI tasks. This model is designed to deliver state-of-the-art accuracy in science, math, and coding while being highly efficient. Notably, it achieves up to three times higher throughput and can run on a single GPU, which lowers computational costs and simplifies deployment for both individual developers and large enterprises. By making this powerful tool openly available, NVIDIA aims to accelerate the development of more capable and reliable AI agents. ...