Liam Price, an amateur mathematician, utilized OpenAI's GPT-5.4 Pro to solve a 60-year-old mathematical conjecture regarding primitive sets, employing a completely new method that experts believe could have broader applications. This achievement highlights the potential of advanced AI models to generate original mathematical insights.
Sources: Hacker News
DeepSeek has released its DSV4 Pro and Flash models, featuring a 1M token context window, new efficient hybrid attention (CSA/HCA), and an MIT license. These open-weight models, built on earlier mHC research, excel in long-context and agentic coding, while DeepSeek also announced plans for Huawei CANN compatibility to reduce NVIDIA dependence.
Sources: Latent Space
OpenAI has launched GPT-5.5, a new iteration of its large language model. This release signifies continued advancements in OpenAI's foundational AI capabilities, potentially offering improved performance and new features for developers and users.
Sources: Simon Willison
OpenAI has launched its new GPT-5.5 large language model through its API, making a significant update available to developers. To ensure optimal performance, the company also published comprehensive guides recommending developers treat GPT-5.5 as a new model family and start with fresh prompt baselines rather than directly porting older configurations.
Sources: Simon Willison
AWS has launched Amazon Quick, an agentic AI workspace designed to unify enterprise knowledge, business intelligence, and workflow automation. Alongside this, AWS introduced the Model Context Protocol (MCP), an open standard for AI agent interoperability. Visier's Workforce AI platform, which provides comprehensive people analytics, is among the first to integrate with Amazon Quick via MCP, creating a unified environment for enhanced decision-making with live workforce data.
Sources: AWS ML Blog
OpenAI has unified its Codex coding model with the main GPT system since GPT-5.4, and the upcoming GPT-5.5 will feature significant improvements in agentic coding and general computer use, eliminating a separate coding-focused model. This move streamlines OpenAI's offerings and enhances the capabilities of its flagship LLM for complex tasks.
Sources: Simon Willison
Nilay Patel's essay, 'The people do not yearn for automation,' argues that the tech industry's 'software brain' perspective, which seeks to automate all aspects of life with AI, alienates the general public by flattening human experience and failing to recognize non-business values.
Sources: Simon Willison
Google's Gemini is rolling out significant updates, integrating advanced AI features into core services like Gmail for email summarization and organization, and Google Maps for optimized errand planning. New multimodal capabilities also enhance home management with tools for cleaning, recipes, and repairs, while also powering third-party virtual home design applications, showcasing a broad expansion of its utility.
Sources: Google AI Blog
Terra API offers a unified platform to connect, normalize, stream, and analyze health data from various siloed sources, enabling AI labs and health companies to build next-generation health intelligence products. The company's vision is to facilitate extreme personalization in health through continuous, real-world health data for AI.
Sources: Hacker News