Chain of News 21/07/2026

**Top Story** The European Commission has published guidelines on transparency obligations for AI systems under Article 50 of the AI Act, marking a significant development in the regulation of artificial intelligence. This move is crucial for developers as it sets a precedent for the transparency and accountability of AI systems, which will have far-reaching implications for the development and deployment of AI models. The guidelines aim to ensure that AI systems are designed and developed with transparency and explainability in mind, which will help build trust in these systems. As AI becomes increasingly pervasive in various aspects of life, the need for transparency and accountability becomes more pressing, and these guidelines are a step in the right direction. The guidelines will likely influence the development of AI systems, and developers should take note of these guidelines to ensure compliance. SOURCES: [2] **AI Models & Research** The Length Value Model is a scalable value pretraining approach for token-level length modeling, which is essential for modern autoregressive models. This model addresses the lack of fine-grained length modeling in existing approaches, which can significantly impact inference cost and reasoning performance. The model's ability to learn token-level length distributions can improve the performance of autoregressive models, making it a valuable tool for developers. Another significant development is the discovery of rater state bias in RLHF preference data, which can affect the accuracy of reinforcement learning models. The audit framework proposed to address this issue can help developers identify and mitigate this bias, leading to more reliable models. Additionally, research has shown that some large language models exhibit consistent risk attitudes, which is a critical dimension to consider when deploying AI systems in high-stakes settings. SOURCES: [1], [4], [5] **Developer Tools & Frameworks** The Copilot AI billing shock has led to the development of meters, caps, and token-saving tools to help developers manage their costs. These tools can provide developers with more control over their expenses and help them optimize their usage of AI-powered development tools. The introduction of these tools is a response to the growing demand for more transparent and predictable pricing models for AI-powered development tools. With these tools, developers can better manage their costs and focus on building innovative applications. Furthermore, the development of lightweight 1D CNNs for affective touch classification in soft plush companions can enable the creation of more intuitive and emotionally intelligent interfaces for socially assistive technologies. SOURCES: [10], [7] **Industry & Business** Anthropic's landmark $1.5B copyright settlement has been approved, settling one case but leaving the broader issue of using copyrighted works to train AI models unresolved. This settlement highlights the ongoing challenges and uncertainties surrounding the use of copyrighted materials in AI model training. The approval of this settlement is a significant development in the AI industry, as it sets a precedent for the resolution of copyright disputes related to AI model training. In another development, Sony Music Entertainment has filed a lawsuit against Udio's AI music generator, accusing it of infringing the copyright of over 30,000 songs. This lawsuit underscores the growing concerns about copyright infringement in the AI music generation space. SOURCES: [8], [9] **Worth Watching** A study has revealed that the more people interact with AI, the more they struggle to recognize errors, a phenomenon that is being referred to as "behaving like ChatGPT." This finding has significant implications for the development of AI systems, as it highlights the need for more transparent and explainable AI models. The study's results suggest that developers should prioritize the design of AI systems that can provide clear and accurate information, rather than relying on users to detect errors. Additionally, the survey on GNN-based link prediction techniques, applications, and challenges provides a comprehensive overview of the current state of graph neural networks, which can be a valuable resource for developers working on link prediction tasks. SOURCES: [3], [6]