Chain of News 25/06/2026

**Top Story** The most significant news of the day is the announcement of a new chip designed for large language model (LLM) inference at scale by OpenAI and Broadcom. This development is crucial as it aims to address the increasing demand for efficient and scalable AI processing. The new chip, named Jalapeño, is OpenAI's first proprietary chip designed to accelerate artificial intelligence. This innovation has significant implications for developers, as it enables the creation of more powerful and efficient AI models. With the ability to process LLMs at scale, developers can now build more complex and sophisticated AI applications, leading to potential breakthroughs in various industries. The collaboration between OpenAI and Broadcom demonstrates the growing importance of hardware-software co-design in the development of AI technologies. **AI Models & Research** The paper "Beyond Shapley: Efficient Computation of Asymmetric Shapley Values" presents a significant advancement in the field of explainability in machine learning models. The authors propose a variant of Shapley values known as Asymmetric Shapley Values (ASV), which enables the incorporation of causal knowledge into feature attribution methods. This research is essential for developers, as it provides a more efficient and effective way to explain the decisions made by complex AI models. Another notable paper is "Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR," which extends reinforcement learning with verifiable rewards (RLVR) to multi-domain reasoning suites. This work has the potential to improve the transferability of AI models across different domains, making them more versatile and applicable to real-world problems. The study "Do vision-language models search like humans?" also deserves attention, as it investigates whether vision-language models search for information in a similar way to humans, providing insights into the decision-making processes of these models. **Developer Tools & Frameworks** The announcement of OpenAI's Jalapeño chip is not only a significant development in the field of AI hardware but also has implications for developer tools and frameworks. With the ability to process LLMs at scale, developers can now build more complex and sophisticated AI applications using frameworks such as TensorFlow or PyTorch. The Jalapeño chip is designed to work seamlessly with OpenAI's existing software stack, making it easier for developers to integrate the new hardware into their workflows. Additionally, the development of more efficient and scalable AI processing capabilities will likely lead to the creation of new tools and frameworks that can take advantage of these advancements. For example, developers may be able to use the Jalapeño chip to build more efficient and accurate natural language processing models, leading to breakthroughs in areas such as language translation or text summarization. **Industry & Business** Anthropic has accused Alibaba of massively copying data from its AI model Claude, highlighting the growing concerns about data privacy and intellectual property in the AI industry. This incident has significant implications for the development of AI models, as it underscores the need for robust data protection and secure collaboration practices. In another development, the US government has blocked the launch of Anthropic's advanced AI models, citing concerns about their potential impact on national security. This decision has sparked alarm in Brussels, with EU officials expressing concerns about the potential consequences of such restrictions on the development of AI technologies. The partnership between OpenAI and Broadcom to develop the Jalapeño chip is also noteworthy, as it demonstrates the growing collaboration between industry leaders to drive innovation in AI hardware and software. **Worth Watching** The paper "The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing" is an interesting read, as it explores the challenges and limitations of autonomous AI systems in healthcare. The authors discuss the need for clinicians to have a "veto" power over AI-driven decisions, highlighting the importance of human oversight and accountability in AI-driven healthcare applications. Another worth-watching development is the study "Elo-Disentangled Player-Style Embeddings for Human Chess via Rating-Conditioned Residual Move Model," which presents a novel approach to learning individual human chess styles using player embeddings. This research has potential applications in areas such as game playing and decision-making, and could lead to new insights into human behavior and strategy. The project "Project Auto-World: Towards Automated Benchmarking of Neural Relational Reasoners" is also worth watching, as it aims to develop automated benchmarking tools for neural relational reasoners, which could lead to significant advancements in areas such as natural language processing and computer vision.