Top Stories: - OpenAI launches GPT-4.1 with improvements in agents, long-context understanding, and coding OpenAI has introduced GPT-4.1, a new model family designed to enhance AI agents, support longer context windows, and improve coding capabilities. This release is part of their ongoing efforts to deliver more powerful and efficient AI models. The Decoder, VentureBeat - OpenAI developing a social network platform similar to X OpenAI is reportedly working on a social network prototype with features centered around ChatGPT’s image generation and social feed, potentially intensifying competition with platforms like X and Meta. The Decoder, The Verge - Anthropic enhances AI assistant Claude with agent-based research and Google Workspace integrationAnthropic’s AI assistant Claude now supports agent-based research workflows and can search users’ entire Google Workspace, boosting productivity and collaboration. The Decoder, VentureBeat - Microsoft launches “Computer Use” feature in Copilot Studio enabling AI agents to interact with desktop appsMicrosoft introduced “Computer Use,” allowing AI agents to operate graphical user interfaces on Microsoft infrastructure for tasks like automated data entry and invoice processing. The Decoder - Google unveils Geospatial Reasoning using generative AI and multiple foundation models Google introduced Geospatial Reasoning, a framework that combines generative AI with geospatial foundation models to provide rapid insights for crisis response, public health, and climate resilience. The system integrates various Google Cloud services and is being tested by organizations like Airbus and Maxar. Google AI Blog - Nvidia starts production of Blackwell AI chips at TSMC’s Arizona plant amid shifting trade landscape Nvidia began production of its new Blackwell AI inference chips at TSMC’s Arizona facility, adapting to export restrictions and trade changes. The Decoder, The Verge, IEEE Spectrum - OpenAI cuts prices for GPT-4.1 models, sparking a competitive AI pricing war OpenAI has significantly lowered prices for GPT-4.1 API usage, prompting other tech companies to adjust their AI service pricing. VentureBeat
Other Stories:
- MIT researchers introduce DISCIPL, a new framework using planner and follower language models for efficient constrained generation and reasoning. MarkTechPost
- Microsoft Research reveals that increasing tokens in LLMs can sometimes degrade reasoning quality, highlighting reasoning fragility in large models. VentureBeat
- Cohere releases Embed 4, a multimodal search model capable of processing large enterprise documents for semantic search applications. The Decoder, VentureBeat
- Google and Pixel phones use AI to interpret dolphin communications in a novel research project. The Decoder
- OpenAI faces a $5.5 billion charge due to export restrictions on AI chips to China. The Verge
- Bank of England warns about potential market instability risks from autonomous AI trading. The Verge
- AI agents capable of using computers autonomously are becoming more common, with Microsoft, OpenAI, and Anthropic developing such technologies. IEEE Spectrum, The Decoder