==The AI industry in 2025 is too noisy, I only picked these 10 sentences as the annual summary== If you, like me, cannot read these dense pieces of information, I will try to extract only one sentence from each of these summary reports and papers. The main thread pieced together by these fragments is that Silicon Valley is committed to evolving AI into "inference agents," and the rise of China's open-source power is breaking the monopoly of this technological evolution. one ️⃣ Karpathy: LLM Annual Review 2025 >We are not 'evolving/cultivating animals', but' summoning ghosts'. two ️⃣ Google DeepMind: Security of Distributed AGI >AGI is not an entity, but a 'state of affairs': a mature, decentralized intelligent agent economy in which humans play a primary role in orchestration and validation. three ️⃣ OpenAI: Frontier Science: Evaluating AI's Ability to Perform Expert level Scientific Tasks >Overall, we have found that cutting-edge AI systems are making rapid progress in solving expert level reasoning problems, especially in self-contained Olympiad questions; But in terms of research style tasks, it is still far from saturated. four ️⃣ OpenAI: The Current Status of Enterprise level Artificial Intelligence in 2025 >The transition from 'requesting output from the model' to 'delegating complex multi-step workflows to the model'. five ️⃣ OpenRouter&a16z: AI Status: Empirical Research on One Trillion Tokens Based on OpenRouter >The focus of the field has shifted from single forward pattern generation to multi-step cautious inference. six ️⃣ How AI is Changing the Way Anthropic Works >Claude is a continuous collaborator, but his use often requires active supervision and verification, especially in high-risk tasks; Instead of handing over tasks that do not require any verification directly. seven ️⃣ DeepSeek V3.2: Advancing the Frontiers of Open Source Large Language Models >The performance of DeepSeek-V3.2 is comparable to GPT-5. eight ️⃣ UC Berkeley/Stanford/IBM Research: Evaluation of Intelligent Agents in Production Environments >Reliability remains the primary development challenge at present, rooted in the difficulty of ensuring and evaluating the correctness of agent behavior. nine ️⃣ Anthropic: AI agent discovers $4.6 million blockchain smart contract vulnerability >Profitable, real-world autonomous vulnerability exploitation is technically feasible. DeepSeek OCR: Context Optical Compression >We are exploring a potential solution: using visual modalities as an efficient compression medium for textual information. Thesis/Research/Report: https://(randomarea. com)