Protesters leak OpenAI's video AI model, Alibaba wants to reason too
📣 AI Weekly will now post on Mondays, bringing you all the AI news from last week to fuel your conversations this week.
what to know for now
🎥 Artists leak OpenAI's Sora model. A group of artists, protesting OpenAI's alleged exploitation in Sora's beta testing, leaked the text-to-video AI model to highlight concerns over unpaid labor and "art washing." The group published Sora's early version on HuggingFace before it was removed, urging OpenAI to adopt transparent, artist-friendly practices. Read more
📉 Intel CEO exits amid AI struggles. Intel CEO Pat Gelsinger has retired as the company faces declining market share in the AI-driven chip sector, with Nvidia dominating the space due to superior GPU capabilities. Intel shares have dropped 61% since Gelsinger's tenure began, while Nvidia’s surged 820%, highlighting Intel's challenges in adapting to AI trends. Read more
🧩 Anthropic’s protocol aims to simplify AI integration. Anthropic has launched the open-source Model Context Protocol (MCP), a standard that connects AI systems to diverse data sources like content repositories and business tools. MCP replaces fragmented integrations with a universal, scalable approach, improving AI’s contextual understanding and response quality. Read more
🧠 Alibaba launches reasoning-focused Marco-o1 AI. Marco-o1, derived from the Qwen2-7B-Instruct model, specializes in complex real-world problem-solving using chain-of-thought fine-tuning, Monte Carlo Tree Search, and reflection mechanisms. It is available on Hugging Face under an Apache 2.0 license, though it is partially open-source and optimized for nuanced tasks like slang translation. Read more
🧪 AI Research of the Week
ROICtrl: Boosting Instance Control for Visual Generation
From Show Lab (National University of Singapore), Meta, MITJake’s Take: ROICtrl, a methodology/adapter defined in this paper, claims to improve AI image generation (diffusion models, specifically) by enabling control over multiple distinct elements of an image by using bounding boxes and captions. In tests, it excels in multi-instance generation with lower computational costs.
This paper could reshape how diffusion models are applied in fine-grained, multi-instance image generation, potentially helping to democratize more precise generations in creative AI tools.
what to know for later
🔍 AI to assist with mass deportations in US. President-elect Trump's administration plans to leverage AI for mass deportations, introducing surveillance tools like drones, robodogs, and biometric systems to enhance enforcement. Experts warn these technologies risk privacy violations, systemic bias, and errors affecting legal residents, while proponents argue AI deregulation could spur innovation. Read more
🛑 Musk challenges OpenAI's profit shift. Elon Musk seeks a court injunction to halt OpenAI's transition to a fully for-profit model, citing antitrust violations and unfair investor agreements. Musk's legal team accuses OpenAI and Microsoft of leveraging sensitive information and market dominance to block competition from startups like xAI. Read more
🎨 Runway AI's Frames previews enhanced image fidelity. The new image-generation model, Frames, focuses on stylistic consistency, enabling users to create variations that adhere to a specific aesthetic. The model will roll out in the Gen-3 Alpha video suite and Runway API, building on Runway's legacy in film and media production. Read more
🚀 Amazon poised to announce Olympus. The Olympus model combines text, image, and video processing, enabling advanced scene search with simple prompts. Expected to debut at AWS re:Invent, it positions Amazon to rival AI leaders. Read more