中文 EN

Bubble's Brain - 2025-12-12

Daily Brief

AI News 2025-12-12

AI Daily Brief

Summary

Disney signed a three-year deal with OpenAI, licensing 200+ iconic characters for AI-generated video and images, and invested $1B.
Zhipu open-sourced four core video-generation technologies to accelerate AGI research and engineering.
Disney also issued a cease-and-desist to Google over alleged copyright infringement in its Gemini/Veo models.
DingTalk 8.1.10 deeply integrates AI into workplace chats and meetings with new smart reply, emoji, and summarization features.
Runway unveiled its general world model GWM-1 to build a reasoning-capable virtual world, alongside major Gen4.5 upgrades.
Google launched a new Gemini Deep Research agent powered by Gemini 3 Pro via a unified Interactions API.
OpenAI confirmed an “adult mode” for ChatGPT is planned for Q1 2026 and released GPT-5.2 to boost everyday professional tasks.

Today’s AI News

  1. Disney and OpenAI strike a deal to license popular characters for video and image creation: Disney and OpenAI announced a three-year partnership that licenses OpenAI to use more than 200 classic characters from Disney, Pixar, Marvel, and Star Wars—primarily for Sora video and ChatGPT image generation. Under the agreement, licensed outputs may be showcased on Disney+, while the likeness and voice of live-action characters are excluded. Disney will also make a $1B equity investment in OpenAI and become a major customer. CEO Bob Iger said the goal is to expand storytelling with generative AI while respecting and protecting creators. The deal is widely seen as part of OpenAI’s efforts to address copyright concerns after launching Sora. At the same time, Disney has issued IP warnings to companies including Google, underscoring its tough stance on copyright protection.

  2. Zhipu’s multimodal open-source week concludes: four core video-generation technologies released: At the end of Zhipu’s Multimodal Open Source Week, the team open-sourced four key technologies for video generation: SCAIL, focused on film-grade character animation and precise control of complex poses; RealVideo, a real-time streaming video system that cuts latency and can output video in about 2–3 seconds; Kaleido, a multi-subject video framework that improves consistency across multiple entities and avoids feature mixing; and SSVAE, which optimizes training to achieve roughly 3× faster convergence at the same quality. Zhipu said the open-sourcing aims to spark community innovation, offer more engineering and research baselines, and advance AGI development.

  3. Disney accuses Google of severe copyright infringement and issues a cease-and-desist: Disney recently sent Google a cease-and-desist notice, alleging that Google’s AI models—such as Gemini and Veo—massively infringed Disney’s copyrighted works, involving characters from franchises like Frozen and Star Wars. Disney argued the models act like a “virtual vending machine,” distributing infringing works for profit, and claimed Google’s copyright protection measures are insufficient. Google responded that the relationship remains constructive and that it will continue discussions, emphasizing that it trains on publicly available web data and offers copyright control mechanisms such as Google-extended. Notably, the notice came shortly before Disney announced its billion-dollar partnership with OpenAI.

  4. DingTalk 8.1.10 major upgrade: AI inside every message, meeting, and recording device: In its latest 8.1.10 release, DingTalk deeply integrates AI into workplace communication. In chat, the new AI Smart Reply generates 4–6 context-aware reply options in different styles and learns user habits over time; DingTalk also adds AI emoji reactions and AI forwarded-message summaries. In meetings and learning scenarios, AI Notes becomes an interactive collaborator with AI Q&A, can extract action items, and supports classroom highlight marking and post-class questions. DingTalk’s voice-entry hardware DingTalk A1 also upgrades with real-time transcription and free multilingual real-time translation; the UI can flip, subtitles can be cast, and transfer/edit speed is improved. Since its September launch, A1 has become a hot-selling product as DingTalk pushes AI deeper into daily work via software–hardware integration.

  5. Runway releases general world model GWM-1 to build a reasoning-capable virtual world: AI video company Runway entered the world model race with GWM-1, aiming to construct a dynamic simulation environment by predicting video pixels frame-by-frame—one that can capture physical rules and temporal evolution. GWM-1 is rolling out via three specialized branches: GWM-Worlds, an interactive app that generates dynamic worlds from text or images; GWM-Robotics, which uses synthetic data to let robots rehearse behaviors; and GWM-Avatars, focused on digital humans with realistic behavioral logic. Runway’s goal is to merge these branches into one unified general world model. Runway also significantly upgraded its video model Gen4.5, adding native audio generation, one-minute multi-shot synthesis, stronger character consistency, and audio editing—signaling AI video generation is moving toward industrial-grade tooling.

  6. Google launches a new Gemini Deep Research agent (Gemini 3 Pro) embeddable via one API: Google introduced a new Gemini Deep Research Agent based on Gemini 3 Pro, now available to developers via the new Interactions API. The agent follows an iterative research workflow: it decomposes goals into sub-questions, runs multiple rounds of search and reading, then synthesizes conclusions instead of producing a single-pass answer. Google says the new version achieves state-of-the-art results on multiple research benchmarks and is already used in industries such as financial services and biotech (e.g., due diligence automation and biomedical literature analysis). The Interactions API provides a unified interface to interact with Gemini models and agents and to manage complex, long-running task lifecycles—simplifying the build-out of AI applications.

  7. OpenAI confirms ChatGPT “adult mode” timeline: expected in Q1 2026: During a GPT-5.2 briefing, OpenAI’s applications CEO Fidji Simo said ChatGPT’s “adult mode” is expected to debut in Q1 2026. Before launch, OpenAI plans to prioritize improvements to age estimation to automatically determine when to apply content restrictions for minors—minimizing false positives for adults while better protecting teenagers. The move follows earlier hints from CEO Sam Altman and signals further personalization in chatbots, alongside higher expectations for content moderation and youth safety.

  8. OpenAI releases GPT-5.2: a next-generation AI assistant for everyday work: OpenAI launched the GPT-5.2 family positioned as “best for everyday professional use,” including Instant, Thinking, and Pro variants. Compared with GPT-5.1, the new models improve accuracy across tasks such as spreadsheets, presentations, coding, and long-document understanding. OpenAI also upgraded its agentic workflows to let ChatGPT take on more complex tasks. Notion and Shopify are among early enterprise testers. GPT-5.2 rolls out first to paid users, and OpenAI plans to sunset GPT-5.1 in roughly three months.

Comments