OpenAI has released updated versions of its GPT models, including GPT 5.4 Mini, GPT 5.4, and GPT 5.4 Pro. Anthropic has also rolled out new iterations of its Claude AI models, specifically Claude Haiku 4.5, Claude Sonnet 4.6, and Claude Opus 4.6, enhancing the performance and features of its conversational AI offerings.
Sources: 2 Covered by Hacker News, Simon Willison
tiny corp, the developer of tinygrad, has launched its tinybox deep learning computers, including models like the red v2 and green v2, with an exa-scale exabox model planned for 2027. The company has also secured funding and is actively hiring software engineers, operations, and hardware staff to further its vision of commoditizing petaflop computing for AI democratization.
Sources: 1 Covered by Hacker News
The Agent Skills Directory (skills.sh) is highlighted as a growing platform for discovering and managing AI agent skills, indicating a significant step towards a modular and interoperable ecosystem for AI functionalities. Microsoft's Azure AI and GitHub Copilot skills show strong adoption in the directory, and Anthropic's skill-creator is a prominent entry.
Sources: via Keep Bookmarks
Simon Willison details his method of using LLMs, primarily Claude Opus 4.6, to generate detailed profiles of Hacker News users by feeding them their last 1,000 public comments. Willison also expresses ongoing alarm about AI security vulnerabilities, particularly the 'lethal trifecta' (access to private data, untrusted input, and ability to take actions), emphasizing the real but often underestimated risk of such attacks.
Sources: via Simon Willison
A software developer used Gemini Pro as a conceptual tutor for LeetCode problems, receiving hints, attack vectors, and real-world examples. The author also developed prompt engineering strategies to optimize the LLM's performance, but observed context degradation after approximately five problems, highlighting a limitation of current LLM context windows.
Sources: via Hacker News
Simon Willison champions 'agentic engineering,' a methodology where coding agents like Claude Code act as productivity multipliers for skilled developers. He argues that LLMs enhance existing expertise rather than replacing programmers and is currently writing a guide on this approach, detailing his mobile-first, 'YOLO mode' workflow for AI-assisted development.
Sources: via Simon Willison
Product management strategies are evolving from stable roadmaps to agile sprints and continuous re-evaluation of features, driven by the rapid pace of AI model advancements. Anthropic's Claude Code on Claude desktop and the Opus 4.6 model are highlighted for significantly lowering the barrier to building and demonstrating AI product prototypes, enabling faster iteration and deployment of new AI features.
Sources: via Keep Bookmarks