[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-ai-weekly-2026-w21-en":3,"article-related-ai-weekly-2026-w21-en":29,"series-industry-5690cc56-568d-4426-a66d-8c3d1296726a":72},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":19,"translated_content":10,"views":20,"is_premium":21,"created_at":22,"updated_at":22,"cover_image":11,"published_at":23,"rewrite_status":24,"rewrite_error":10,"rewritten_from_id":10,"slug":25,"category":26,"related_article_id":27,"status":28,"google_indexed_at":10,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":10,"topic_cluster_id":10,"embedding":10,"is_canonical_seed":21},"5690cc56-568d-4426-a66d-8c3d1296726a","AI Weekly: 2026-05-11 ~ 2026-05-18","\u003Cp>This week’s AI news had a clear theme: the field is moving from demos to systems that have to justify their cost, reliability, and operational fit. The strongest stories were less about shiny new models and more about how teams are actually shipping, scaling, and governing AI in production.\u003C\u002Fp>\u003Cp>That shift shows up across research, tools, and industry debates. From test-time scaling experiments to open-source RAG systems, the gap between “can it work?” and “can we run it every day?” is getting harder to ignore.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fautotts-llms-discover-test-time-scaling-en\">AutoTTS lets LLMs discover test-time scaling\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>AutoTTS reframes test-time scaling as an environment search problem, which is a neat way of saying models can learn how to think more efficiently instead of brute-forcing every answer. That matters because test-time compute has become one of the biggest hidden costs in serious LLM use, especially when chains of reasoning get long.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779077043425-08ip.png\" alt=\"AI Weekly: 2026-05-11 ~ 2026-05-18\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The key change is not just better performance, but a method for finding cheaper reasoning strategies automatically. If this holds up outside controlled experiments, it could reduce the need for hand-tuned inference tricks and make advanced reasoning more practical for teams watching latency and token bills.\u003C\u002Fp>\u003Cp>Researchers and infrastructure teams should both care. For model builders, it opens a path toward more adaptive inference policies; for product teams, it may mean better answers without paying for the most expensive path every time.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fragflow-open-source-rag-agent-engine-en\">RAGFlow adds agents to open-source RAG\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>RAGFlow’s update is a reminder that retrieval alone is no longer enough for many enterprise use cases. By adding agent features on top of open-source RAG, plus Docker self-hosting and support for newer models like GPT-5 and Gemini 3 Pro, it moves closer to a full application layer rather than a narrow search wrapper.\u003C\u002Fp>\u003Cp>This matters because many teams have already learned that “chat with your docs” is the easy part. The hard part is orchestration across retrieval, tool use, permissions, and model choice, and RAGFlow is trying to package that into something deployable without locking teams into a single vendor.\u003C\u002Fp>\u003Cp>The affected audience is broad: startups building internal copilots, enterprises that need on-prem or controlled deployment, and developers who want a more complete open-source stack. It also raises the bar for competing RAG tools, which now need to explain why they stop at retrieval.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fhow-to-use-openai-sora-in-2026-en\">How to Use OpenAI Sora in 2026\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>Guides like this matter because video generation is finally crossing from curiosity into workflow. Sora is no longer being discussed only as a wow demo; the practical question is how to prompt, refine, and iterate without wasting time on unusable clips.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779077040695-2i6t.png\" alt=\"AI Weekly: 2026-05-11 ~ 2026-05-18\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The update signals a broader change in AI media: the skill is shifting from “can you make a video?” to “can you direct the model well enough to get a usable shot?” That changes the value of creative teams, marketers, and solo builders who need fast visual output but do not have a full production pipeline.\u003C\u002Fp>\u003Cp>Anyone experimenting with AI video should pay attention, especially teams that care about brand consistency and revision speed. The real story is not that video generation exists, but that the workflow around it is getting more structured and less mysterious.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fhow-to-run-hermes-agent-on-discord-en\">How to Run Hermes Agent on Discord\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>Discord bots remain one of the fastest ways to turn an AI agent into something people actually use, and Hermes Agent fits that pattern well. The setup details around Discord intents and OpenClaw Launch deployment are unglamorous, but they are exactly what separates a toy bot from a service people can rely on.\u003C\u002Fp>\u003Cp>This story matters because community platforms are becoming a proving ground for agents. If an agent can operate in a busy Discord server, handle permissions correctly, and stay online through deployment friction, it has a better chance of surviving in more serious environments later.\u003C\u002Fp>\u003Cp>Builders should care most. The affected groups are indie developers, community operators, and teams testing agent workflows before moving them into internal tools. In practice, Discord is still one of the best places to see whether an agent is actually useful or just conversational.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fwhy-container-design-patterns-matter-more-than-orchestration-en\">Why container design patterns matter more than orchestration\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>This piece pushes back on a common habit in infrastructure discussions: treating orchestration as the main event. The argument is that container design patterns are the real unit of distributed-systems thinking, because they shape how services fail, recover, and interact long before a scheduler gets involved.\u003C\u002Fp>\u003Cp>That point lands especially well in AI infrastructure, where teams often focus on deployment tools while ignoring the structure of the workloads themselves. If your container boundaries are messy, orchestration will not save you; it will just automate the mess faster.\u003C\u002Fp>\u003Cp>Platform teams, SREs, and AI engineers should take this seriously. As more AI workloads mix model calls, retrieval, background jobs, and stateful components, the design of the container matters more than the brand name of the orchestrator.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fwhy-ai-agents-should-maintain-your-wiki-en\">Why AI agents should maintain your wiki, not answer your questions\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>This is one of the more practical agent arguments of the week. Instead of asking agents to answer the same questions over and over, the better use case is to have them maintain a living wiki that captures what the organization already knows and keeps it current.\u003C\u002Fp>\u003Cp>That matters because repeated Q&A is a poor use of agent time and a weak way to build organizational memory. A maintained wiki creates a durable source of truth, reduces duplicated answers, and gives humans a place to review and correct the record.\u003C\u002Fp>\u003Cp>The people most affected are knowledge workers, support teams, and internal tooling owners. If agents are going to justify their cost, they need to do more than respond; they need to improve the information base that everyone else depends on.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fus-should-keep-frontier-ai-out-of-china-en\">Why the U.S. should keep frontier AI out of China\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>This policy argument is blunt: strategic risk should outweigh commercial temptation when it comes to frontier AI access. The core claim is that advanced model capabilities are too important to treat like ordinary exportable software, especially when the geopolitical stakes are this high.\u003C\u002Fp>\u003Cp>Whether one agrees or not, the story reflects how AI policy is hardening around national security rather than pure market logic. Frontier models are no longer seen as just powerful products; they are increasingly treated as strategic assets with military, intelligence, and industrial implications.\u003C\u002Fp>\u003Cp>The affected parties include model providers, policymakers, and global enterprises operating across borders. Even firms that are far from the policy debate will feel the impact through compliance pressure, licensing rules, and changes in where advanced AI can be deployed.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fwhy-buns-zig-to-rust-experiment-is-right-en\">Why Bun’s Zig-to-Rust experiment is the right move\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>Bun’s experiment is interesting because it treats language choice as an engineering question rather than a branding decision. Moving from Zig to Rust is not a small shift, but it is the kind of move that can expose real trade-offs in speed, safety, and developer effort.\u003C\u002Fp>\u003Cp>That matters in a week when AI tooling and infrastructure are both under pressure to become more dependable. If the runtime and tooling layer is fragile, every AI app built on top inherits that fragility, no matter how polished the model layer looks.\u003C\u002Fp>\u003Cp>Developers and infrastructure teams will be watching closely. The result could influence how other fast-moving toolmakers think about implementation risk, especially when performance claims need to survive contact with production.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fprompt-engineering-jobs-2026-worth-it-en\">Prompt Engineering Jobs in 2026: Still Worth It?\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>The answer here is basically yes, but with an important caveat: prompt engineering is no longer a standalone destination for most people. The best roles now sit inside product, engineering, and operations work where prompting is one skill among several.\u003C\u002Fp>\u003Cp>This reflects a broader maturation of the market. As models improve, the value shifts away from writing clever prompts and toward designing systems, evaluating outputs, and making AI features reliable enough for users who do not care how the prompt was written.\u003C\u002Fp>\u003Cp>Job seekers and hiring managers should both read this carefully. The market is still real, but it rewards people who can connect prompting to product outcomes, evaluation, and deployment rather than treating it as a narrow specialty.\u003C\u002Fp>\u003Ch2>\u003Ca href=\"\u002Fnews\u002Fmlops-in-2026-architecture-strategy-guide-en\">MLOps in 2026: Architecture and Strategy Guide\u003C\u002Fa>\u003C\u002Fh2>\u003Cp>MLOps in 2026 is less about model hosting and more about governance, cost control, and the convergence of LLMOps with broader enterprise operations. That is a healthy correction: after years of pilot-heavy enthusiasm, companies are now asking what it takes to run AI as part of the business, not as a lab demo.\u003C\u002Fp>\u003Cp>The article’s value is in spelling out that the stack is changing. Teams need observability, policy controls, evaluation, and cost discipline, because model behavior alone is not what makes production AI succeed or fail.\u003C\u002Fp>\u003Cp>This affects enterprise architects, ML engineers, and operations leaders. The organizations that get this right will be the ones that treat AI as an operational system with budgets, guardrails, and ownership, not as a one-time model rollout.\u003C\u002Fp>\u003Ch2>What to watch next week\u003C\u002Fh2>\u003Cul>\u003Cli>More evidence on whether test-time scaling methods can cut inference cost without hurting answer quality.\u003C\u002Fli>\u003Cli>Whether open-source RAG systems keep absorbing agent features, or whether the stack starts to split again.\u003C\u002Fli>\u003Cli>Fresh signs that AI video tools are moving from novelty workflows into repeatable production use.\u003C\u002Fli>\u003Cli>How enterprise teams are handling governance, evaluation, and cost as MLOps and LLMOps continue to merge.\u003C\u002Fli>\u003C\u002Ful>","This week in AI: AutoTTS trims test-time cost, RAGFlow adds agents, Sora gets practical, and MLOps keeps shifting toward governance.","oracore.dev","https:\u002F\u002Foracore.dev\u002Fnews\u002Fai-weekly-2026-w21-en",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779077043425-08ip.png",[13,14,15,16,17,18],"AI Weekly","AI news","artificial intelligence","RAG","MLOps","AI agents","en",0,false,"2026-05-18T04:00:39.784477+00:00","2026-05-18T04:00:39.707+00:00","done","ai-weekly-2026-w21-en","industry","15fa8a17-81dc-43cb-9a2d-774364314e79","published",{"tags":30,"relatedLang":31,"relatedPosts":35},[],{"id":27,"slug":32,"title":33,"language":34},"ai-weekly-2026-w21-zh","AI 週報：2026-05-11 ~ 2026-05-18","zh",[36,42,48,54,60,66],{"id":37,"slug":38,"title":39,"cover_image":40,"image_url":40,"created_at":41,"category":26},"00c79ad1-a351-40b7-9f36-0c590f90b935","why-footwear-news-still-matters-sneaker-market-en","Why Footwear News still matters in a sneaker-first market","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779082430656-xoub.png","2026-05-18T05:33:24.747061+00:00",{"id":43,"slug":44,"title":45,"cover_image":46,"image_url":46,"created_at":47,"category":26},"9b888214-986c-4269-97cd-ff52e215289f","giants-news-roster-schedule-injuries-heavy-en","Giants News, Roster, Schedule and Injuries","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779081842606-pc9f.png","2026-05-18T05:23:38.784392+00:00",{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":26},"715d7b09-d772-4278-8c38-d1945db900c5","utah-jazz-2026-roster-injury-report-stats-en","Utah Jazz enter 2026 with a thin roster","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779081246920-w4zd.png","2026-05-18T05:13:38.576486+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":26},"26e51937-1657-4ac4-bcff-f0425ceafee7","why-minimax-matters-more-as-a-consumer-ai-company-en","Why MiniMax Matters More as a Consumer AI Company Than a Model Lab","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779078234363-8yhw.png","2026-05-18T04:23:24.926213+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":26},"852ba92c-da65-4985-9941-932719583d03","why-distributed-computing-is-the-default-en","Why Distributed Computing Is the Default, Not the Exception","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779075830914-h1ov.png","2026-05-18T03:43:25.148261+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":26},"823930b5-566e-4643-b1eb-2732c6f01dbf","community-resistance-will-reshape-ai-data-center-expansion-en","Why community resistance will reshape AI data center expansion","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779074628087-4e71.png","2026-05-18T03:23:20.615889+00:00",[73,78,83,88,93,98,103,108,113,118],{"id":74,"slug":75,"title":76,"created_at":77},"d35a1bd9-e709-412e-a2df-392df1dc572a","ai-impact-2026-developments-market-en","AI's Impact in 2026: Key Developments and Market Shifts","2026-03-25T16:20:33.205823+00:00",{"id":79,"slug":80,"title":81,"created_at":82},"5ed27921-5fd6-492e-8c59-78393bf37710","trumps-ai-legislative-framework-en","Trump's AI Legislative Framework: What's Inside?","2026-03-25T16:22:20.005325+00:00",{"id":84,"slug":85,"title":86,"created_at":87},"e454a642-f03c-4794-b185-5f651aebbaca","nvidia-gtc-2026-key-highlights-innovations-en","NVIDIA GTC 2026: Key Highlights and Innovations","2026-03-25T16:22:47.882615+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"0ebb5b16-774a-4922-945d-5f2ce1df5a6d","claude-usage-diversifies-learning-curves-en","Claude Usage Diversifies, Learning Curves Emerge","2026-03-25T16:25:50.770376+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"69934e86-2fc5-4280-8223-7b917a48ace8","openclaw-ai-commoditization-concerns-en","OpenClaw's Rise Raises Concerns of AI Model Commoditization","2026-03-25T16:26:30.582047+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"b4b2575b-2ac8-46b2-b90e-ab1d7c060797","google-gemini-ai-rollout-2026-en","Google's Gemini AI Rollout Extended to 2026","2026-03-25T16:28:14.808842+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"6e18bc65-42ae-4ad0-b564-67d7f66b979e","meta-llama4-fabricated-results-scandal-en","Meta's Llama 4 Scandal: Fabricated AI Test Results Unveiled","2026-03-25T16:29:15.482836+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"bf888e9d-08be-4f47-996c-7b24b5ab3500","accenture-mistral-ai-deployment-en","Accenture and Mistral AI Team Up for AI Deployment","2026-03-25T16:31:01.894655+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"5382b536-fad2-49c6-ac85-9eb2bae49f35","mistral-ai-high-stakes-2026-en","Mistral AI: Facing High Stakes in 2026","2026-03-25T16:31:39.941974+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"9da3d2d6-b669-4971-ba1d-17fdb3548ed5","cursors-meteoric-rise-pressures-en","Cursor's Meteoric Rise Faces Industry Pressures","2026-03-25T16:32:21.899217+00:00"]