#genai

Toward More Controllable AI Video Editing: An Early Research Exploration at Netflix

Netflix giới thiệu hai mô hình chỉnh sửa video AI giai đoạn đầu là Vera và VOID. Vera sử dụng mô hình diffusion phân lớp, chỉ tái tạo vùng chỉnh sửa (kèm alpha matte) thay vì toàn bộ clip, bảo toàn nội dung chưa chỉnh sửa. VOID chuyên xóa vật thể trong video với kỹ thuật inpainting hợp lý vật lý, tái tạo cảnh thực tế khi vật thể bị loại bỏ. Cả hai mô hình đều vượt trội so với các phương pháp hiện có trong nghiên cứu.

Lập trình viên muốn phát triển các giải pháp AI tiên tiến trong xử lý video nên tham khảo để hiểu cách thiết kế mô hình hiệu quả như Vera và VOID, từ kiến trúc đặc biệt đến kỹ thuật điều khiển chi tiết để nâng cao chất lượng và tính khả thi của các ứng dụng AI video trong tương lai.

MIT in the media: Exploring how curiosity-driven science is an essential ingredient in America’s success

Bài viết trên tạp chí Scientific American giới thiệu vai trò quan trọng của nghiên cứu khoa học định hướng tò mò (curiosity-driven research) trong thành công của Mỹ, với sự đóng góp của các giảng viên, sinh viên và cựu sinh viên MIT. Các chuyên gia nhấn mạnh nhu cầu đầu tư công bền vững vào khoa học cơ bản, đồng thời lo ngại về sự bất ổn trong tài trợ liên bang, rào cản nhập cư và suy giảm niềm tin vào chuyên gia, dù vẫn lạc quan nhờ lịch sử đổi mới của Mỹ.

Lập trình viên nên đọc bài này để hiểu cách các nhà khoa học tự do khám phá (curiosity-driven) không chỉ thúc đẩy tiến bộ khoa học mà còn tạo ra những công nghệ đột phá—như AI, sinh học tổng hợp và năng lượng tái tạo—đã định hình nền tảng cho các ứng dụng tương lai của phần mềm và hệ thống thông minh.

Microsoft raises price of Xbox consoles sunsets 2TB Xbox Series X

Microsoft is raising Xbox Series X|S prices starting August 1, 2026, with the 512GB Xbox Series S increasing by $100 to $499.99 and 1TB models rising $150, bringing the Xbox Series X to $749.99. The 2TB Xbox Series X will be discontinued. Microsoft directly attributes the hikes to a global memory hardware shortage caused by generative AI companies securing exclusive deals with DRAM manufacturers, driving component costs up more than 2.5x with further doubling expected by fall 2027. Valve's upcoming Steam Machine faces similar pricing pressure. Xbox CEO Asha Sharma has warned investors the shortage may force a rethink of the next-gen Project Helix console, alongside a broader business reset that includes reported layoffs. The article notes the irony that Microsoft itself is a major driver of generative AI adoption.

Unconventional AI releases its first model, built on an oscillator architecture that could cut AI power use by 1000x

Unconventional AI, founded by former Databricks AI chief Naveen Rao, has released Un-0, an image generation model running on a software simulation of a novel oscillator-based computing architecture. Unlike conventional digital logic using transistors, the approach uses coupled ring oscillators to encode and process information through physics, with Rao claiming it could reduce AI power consumption by 1000x. The model performs comparably to established diffusion models like Stable Diffusion, but currently runs only on a simulated version of hardware that doesn't yet exist. The company, backed by $475M in seed funding at a $4.5B valuation from Lightspeed, a16z, and Sequoia, plans to release chip schematics soon and eventually operate as a compute provider. The 1000x efficiency claim remains theoretical until physical hardware is built and validated.

Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x

Unconventional AI, led by former Databricks AI chief Naveen Rao, has released Un0, an image-generation model built on a software simulation of a novel oscillator-based computing architecture. The company claims this architecture could reduce AI inference power consumption by up to 1,000x compared to conventional chips. Un0 performs comparably to state-of-the-art diffusion models like Stable Diffusion, serving as a proof-of-concept for the new architecture. The company plans to release actual chip schematics soon and eventually build a full inference stack, positioning itself as a compute provider running at a fraction of current energy costs.

Magic: The Gathering Arena developers have successfully unionized

Magic: The Gathering Arena developers at Wizards of the Coast and Hasbro have successfully unionized under the United Wizards of the Coast-CWA (UWOTC-CWA) banner following an NLRB election on June 2, 2026. Workers sought protections around generative AI use, layoffs, and compensation transparency. After Wizards of the Coast declined voluntary recognition and reportedly discouraged unionization efforts, the majority voted in favor. The union now aims to ratify a contract and hopes to inspire other game industry workers to organize.

Connecting Figma and Weave

Figma is integrating its Weave AI creative platform more deeply into the Figma design canvas. Three key updates are announced: 20+ AI image tools (style transfers, product shoots, material extraction) are now accessible directly from Figma Design's left panel as pre-built Weave workflows; Weave workflows can now be published and discovered on the Figma Community, letting designers share reusable creative pipelines; and a forthcoming Figma node in Weave will allow designers to paste Figma frames directly into Weave workflows, with real-time sync between design edits and the creative pipeline. Weave tools are in open beta and free during the beta period, with Figma AI credits required once generally available.

Swiss study finds fewer job ads for career starters as AI reshapes entry-level work

A Jobs.ch analysis of 7.3 million Swiss job advertisements finds that entry-level postings in 2025 were nearly a third lower than the pre-generative-AI baseline (2019–2022). The decline is steepest in AI-exposed white-collar sectors — administration, finance, HR, marketing, IT — where junior role share fell 16% while senior role share rose 26%. The study stops short of attributing the shift solely to AI, noting a cooling economy as a confounding factor, but the pattern aligns with broader European research. The core concern is that routine entry-level tasks — drafting, data entry, first-pass analysis — are being absorbed by AI tools, removing the traditional on-ramp through which newcomers develop professional judgment and skills.

ByteDance unveils Seedance 2.5, a 30-second native 4K AI video model that accepts 50 reference inputs

ByteDance unveiled Seedance 2.5 at its Volcano Engine FORCE conference in Beijing, a video generation model capable of producing 30-second native 4K clips from up to 50 multimodal reference inputs including images, audio, 3D models, and style references. The model features native 4K output (not upscaled), 10-bit colour depth, 20% better prompt adherence, and co-processed audio synchronized with visuals. An enterprise beta is live with public launch targeted for early July. The announcement follows copyright disputes with major Hollywood studios that stalled the global rollout of its predecessor. Competitively, it significantly outpaces Google's Veo 3.1 on reference input capacity (50 vs. 3), while OpenAI's Sora has shut down. ByteDance's path to US markets remains uncertain due to unresolved copyright conflicts.

SageMaker Notebook Instances now support G6e instance types

AWS announces general availability of Amazon EC2 G6e instances on SageMaker notebook instances. Powered by up to 8 NVIDIA L40S Tensor Core GPUs with 48 GB memory per GPU and AMD EPYC processors, G6e instances offer up to 2.5x better performance than G5 instances. They support interactive model training use cases including generative AI fine-tuning and deployment of LLMs with up to 13B parameters, as well as diffusion models for image, video, and audio generation. Available in multiple AWS regions including US East, US West, Asia Pacific, Middle East, and Europe.

Automated Reasoning checks in Amazon Bedrock Guardrails add new policy refinement workflows

AWS has added two new automated policy refinement workflows to Automated Reasoning checks in Amazon Bedrock Guardrails. The first, an iterative policy improvement workflow, lets customers run refinement passes against natural language tests to automatically deduce and apply policy changes. The second, an ambiguity reduction workflow, automatically refines variable descriptions and type definitions to reduce ambiguous translation results. Both workflows are accessible via the Amazon Bedrock APIs and the AWS Management Console, and are available in all regions where Automated Reasoning checks are supported.

EA: AI is delivering "faster prototyping" and a "real rise in creativity" in its studios

EA's president of enterprise development Laura Miele claims AI is delivering faster prototyping and a rise in creativity at EA's studios by removing tedious tasks from development workflows. She made the remarks at The Game Business Live during Summer Game Fest 2026. EA CEO Andrew Wilson previously stated that over 50% of EA's development processes would be positively impacted by generative AI. The comments come amid broader industry controversy, with surveys showing more than four-fifths of games industry workers finding AI-generated content unacceptable. EA is also currently being acquired by a consortium including Saudi Arabia's Public Investment Fund for $55 billion.

Video Game Lawyer On Legal Concerns Surrounding UE6's AI Tools

Video game lawyer Haley MacLean, speaking on the MinnMax podcast, outlines key legal concerns around Unreal Engine 6's planned native generative AI integration. The core issue is that AI-generated content cannot be copyrighted under current legal systems worldwide, which require human authorship. This creates risks around reusing AI-generated assets across multiple games and opens the door for bad-faith publishers to exploit developers — for example, by withdrawing from development contracts when AI-generated assets are involved and cannot be properly licensed.

Vibecoding Malware Fuels a New Wave Of Stealth Attacks

Cybersecurity researchers at University College Cork have found that AI-assisted malware creation, dubbed 'vibeware,' poses a growing threat by generating functionally identical but structurally varied malicious code that evades traditional static detection methods like hash matching and YARA rules. The UCC team demonstrated that malware targeting Linux systems could be crafted with as few as two prompts using tools like Cursor, which did not restrict the requests. Experts from McAfee, Bitdefender, and ProjectDiscovery warn that LLMs lower the barrier to entry for threat actors, enabling even non-coders to build malware. The consensus is that defenders must shift to behavioral and AI-driven detection strategies. Google's discovery of PROMPTFLUX — malware that calls the Gemini API at runtime to rewrite its own code — illustrates how adaptive AI-generated malware could become.