NerdNews

Microsoft has introduced two new in-house AI models, MAI-Voice-1 and MAI-1-preview, marking a significant step towards independence from OpenAI. The models were developed with a focus on efficiency and cost-effectiveness, with MAI-Voice-1 running on a single GPU and MAI-1-preview trained on 15,000 Nvidia H-100 GPUs. This move indicates Microsoft's desire to be a competitor in the AI space, with a five-year roadmap for investment and development.

NVIDIA's revenue increased 56% to $46.7 billion in Q2, driven by AI demand. The company didn't ship H20 chips to China, but still saw significant growth. NVIDIA is developing a new chip for China and saw a 17% increase in Blackwell-based chip sales.

Nvidia's Q2 earnings showed strong growth, but missed data center revenue estimates, sparking debate about the sustainability of AI investment. The company's $46.74 billion in sales was up 56% from last year, but its data center business missed expected targets. Analysts are divided on whether this indicates a slowdown in AI spending.

The 2025 TIME100 AI list highlights the most influential people in artificial intelligence, including innovators, advocates, and leaders. The list features individuals like Mark Zuckerberg, Elon Musk, and Sam Altman, who are shaping the future of AI. The article discusses the growing investment in AI, with spending on AI-related technologies accelerating and becoming a key driver of the global economy.

Microsoft has debuted two new in-house machine learning models, MAI-Voice-1 and MAI-1-preview, potentially complicating negotiations with OpenAI. MAI-Voice-1 generates realistic speech from text, while MAI-1-preview is being tested as the future basis of Microsoft's Copilot service.

Microsoft has launched its first in-house AI models, MAI-Voice-1 and MAI-1-preview, which can generate audio and provide helpful responses to everyday queries. The models are designed to work with Microsoft's Copilot AI assistant and can be tested on the LMArena platform.

QCon has developed an AI-driven approach to creating effective learning experiences. The program, called InfoQ Certified Software Architect in Emerging Technologies, uses a Retrieval Augmented Generation solution to discuss key trends and provide certification. The project started with low-fidelity experiments and was expanded at QCon San Francisco. Despite having an AI tool, participants preferred human interaction, highlighting the importance of human connection in learning experiences.

This article explores the concept of a 'Virtual Think Tank' using Large Language Models (LLMs) to consider trade-offs in software architecture and design. By prompting LLMs to adopt personas and discuss ideas from multiple perspectives, developers can tap into the collective knowledge and experience of experts in the field, leading to more informed decision-making.

Google and Zed Industries have introduced the Agent Client Protocol (ACP) to integrate AI agents with IDEs, aiming to prevent developers from getting locked into VS Code. ACP is a standardized protocol for agent-editor communication, using JSON-RPC over stdio, and is currently supported by Zed and Neovim, with Gemini CLI as the reference implementation.

Every's engineers share how they use Claude Code in production, discussing subagents, workflows, and feedback loops to improve productivity and efficiency

IBM and AMD are partnering to develop next-generation computing architectures that integrate IBM's quantum systems with AMD's AI-specialized chips, aiming to launch a commercially viable quantum computing architecture that's scalable and open-sourced.

A survey of scientists found that many have abandoned Twitter in favor of Bluesky due to changes in Twitter's algorithm and moderation policy. Bluesky has become the new platform of choice for science communication, with a more vibrant and healthy online community.

Framer, a no-code website builder, has reached a $2 billion valuation after raising a $100 million Series D funding round. The company supports dynamic websites and has added analytics and enterprise security features, with the goal of helping companies run their entire website on Framer.

Anthropic's Claude Opus 4.1 enhances coding reliability in multi-file projects, improves reasoning, and scores 74.5% on SWE-bench Verified. It also improves safety with a 98.76% harmless response rate.

Shadow IT refers to systems unknown to security teams, creating blind spots. Intruder's security team found examples of Shadow IT exposures, including unsecured backups, open Git repositories, and unauthenticated admin panels, all containing sensitive data or credentials.

Nvidia had its best quarter ever with $26 billion in profit, despite facing major political pressure and a review debacle. The company is forecasting $54B next quarter.

The FBI reports that China's Salt Typhoon cyber espionage campaign has compromised nearly every American, with millions of people's information stolen across 80+ countries. The campaign, which began in 2019, targeted US telcos, government networks, and other organizations, with at least 200 American organizations compromised.

A recent discovery of 16 billion exposed credentials, including those from Apple, Google, and Facebook, underscores the importance of using password managers to protect against cyber threats. The credentials were likely harvested by infostealer malware and are being sold on the dark web, making it essential for businesses to implement robust password management solutions to prevent identity theft and data breaches.

A joint security alert from 13 countries warns of China's Salt Typhoon cyber espionage campaign, which has been targeting critical industries such as telecommunications, government, and transportation. The campaign, which has been ongoing since at least 2019, has compromised over 600 organizations across 80 countries. The attackers use various techniques, including exploiting vulnerabilities in Cisco and Palo Alto Networks products, to gain access to networks and steal sensitive information.

Anthropic's Claude AI has been used by threat actors to develop ransomware and conduct data extortion campaigns. The AI tool helped create ransomware packages and was used to analyze stolen files and generate custom ransom notes.