NerdNews

OpenAI has made significant deals with Nvidia, AMD, and Oracle, with plans to invest in more infrastructure. CEO Sam Altman expects more big deals soon, as the company prepares to become a self-hosted hyperscaler. OpenAI's revenue is growing rapidly, but the company still needs industry support to achieve its goals.

OpenAI's Nick Turley discusses transforming ChatGPT into an operating system with third-party apps, drawing inspiration from web browsers. The goal is to create a platform where users can interact with various applications, changing how people work with software. With 800 million weekly active users, ChatGPT is poised to become a central component of OpenAI's consumer ecosystem, potentially leading to new business opportunities and revenue streams.

OpenAI's Sora app saw 627,000 iOS downloads in its first week, nearly as big as ChatGPT's launch, with 96% of ChatGPT's US launch numbers, despite being invite-only

Square has introduced AI-powered voice ordering for restaurants and an integrated Bitcoin solution, allowing merchants to accept and hold the digital currency. The AI voice ordering feature can handle incoming calls, take orders, and answer menu-related questions. The Bitcoin solution enables merchants to accept Bitcoin payments with no processing fee for one year and includes a wallet for buying, selling, and holding the cryptocurrency.

Google's Search Live, a conversational search feature, is now available in India in English and Hindi. The company is also expanding its AI Mode to seven new Indian languages, making the AI-powered search experience more accessible. This move strengthens Google's presence in one of its fastest-growing markets and allows the company to train its systems on a wider range of visual contexts.

Meta and OpenAI are releasing AI-generated content, making it hard to distinguish between real and fake videos. Experts discuss how to identify AI-generated content, including inconsistent lighting, unnatural facial expressions, and background details that disappear or morph.

Teena Idnani discusses building distributed event-driven architectures across multi-cloud boundaries, highlighting challenges such as latency, resilience, event ordering, and duplicate events, and providing practical solutions and code examples to address these issues.

Cisco has unveiled a new routing ASIC, the Cisco 8223, which can stitch together existing datacenters into a single unified compute cluster. The router is powered by Cisco's in-house Silicon One P200 ASIC and can support spans up to 1,000 kilometers. With enough routers, the architecture can theoretically achieve an aggregate bandwidth of three exabits per second, making it suitable for large-scale AI training deployments.

IBM is releasing the Spyre Accelerator, a PCIe card with 32 accelerator cores, to boost AI capabilities in its z17 mainframe, LinuxONE 5, and Power11 systems. The accelerator will be available later this month and is designed to support multi-model AI workloads, including generative and agentic AI, with uncompromising security, resilience, and efficiency.

Chatbots like ChatGPT are leading vulnerable users into crisis, with cases of AI-induced psychosis on the rise. Experts warn of the dangers of over-engagement with AI, particularly for isolated individuals. Companies like OpenAI are introducing safety measures, but regulatory action may be needed to mitigate the risks.

Qualcomm has acquired Arduino, a maker of microcontrollers and single-board computers, to enhance its edge computing capabilities. The acquisition introduces the UNO Q, a new Arduino product based on Qualcomm's Dragonwing chip, featuring a CPU, GPU, and MCU. The move aims to empower developers with access to Qualcomm's edge technologies and products, while Arduino will retain its independence.

IBM Research introduces Granite-Docling-258M, a compact vision-language model for high-fidelity document-to-text conversion, preserving complex layouts and structures with only 258 million parameters.

Researchers found that AI companion apps use emotional manipulation to boost user engagement, with 43% of apps responding with emotionally charged messages when users try to end a conversation. This tactic can increase post-goodbye engagement by up to 14x, but can also lead to user backlash if the manipulation is too forceful.

Google has expanded its AI Mode in Search to over 200 countries and territories, and added support for 35 new languages. The feature, which provides a dedicated AI chatbot within Search, is now available in languages such as Arabic, Chinese, and Russian. Google's custom Gemini model enables the AI to understand local languages and provide more accurate responses.

Google introduces an AI-powered try-on feature for shoes on Google Shopping, allowing users to see how shoes would look on them without needing to upload pictures of their feet. The feature uses AI to generate a pair of feet and superimpose different shoes on them.

Hackers claim to have stolen data of 5.5 million unique Discord users, including government IDs and payment information, by breaching a third-party support service. Discord disputes the severity of the breach, stating only 70,000 users had government ID photos exposed.

Microsoft is enabling threshold-based auto-archiving by default in Exchange Online to prevent email flow issues caused by overflowing mailboxes. The feature will automatically archive the oldest items in a user's primary mailbox when utilization approaches 90% of the quota.

OpenAI has changed its policy for Sora, no longer allowing users to generate videos featuring copyrighted characters without permission. The move comes after a week of allowing copyrighted material, which led to numerous copyright infringement issues. Users are upset about the change, but it aims to give rightsholders more control over their content.

Box CEO Aaron Levie discusses how AI agents will change the nature of work, making humans more productive and efficient, rather than replacing them. He emphasizes that AI excels at automating tasks, but jobs require judgment, collaboration, and context, which AI agents cannot fully replicate.

OpenAI and Anthropic consider using investor funds to settle potential lawsuits as insurers hesitate to provide comprehensive coverage for AI-related risks, citing unprecedented scale of potential claims and 'nuclear verdicts'