Alibaba Open Sources AI Video/Image Model Wan 2.1

Also, Simon Willison’s take on Claude 3.7 Sonnet’s “hybrid reasoning”

Author

Jack Lajoie
February 26, 2025 • read time ~ 10 minutes

⚡️ Headlines

🤖 AI

Apple commits to over $500 billion U.S. investment in next four years - Apple plans to expand teams and facilities across multiple states, including a new factory in Texas, doubling its Advanced Manufacturing Fund, and accelerating investments in AI and silicon engineering. [Apple Newsroom]

Edera introduces cloud security tech to enhance AI protection - Founded by an all-female team, Edera aims to improve cloud workload isolation, addressing vulnerabilities in shared cloud environments, particularly for AI workloads relying on GPUs. [WIRED]

DeepSeek accelerates launch of new AI model amid China's AI push - The Chinese startup plans to release its R2 AI model earlier than planned, aiming to enhance coding abilities and expand reasoning capabilities beyond English. [Reuters]

Anthropic finalizing $3.5 billion funding round - The AI startup is set to conclude a significant funding round, valuing the company at $61.5 billion, to develop more advanced AI models. [The Wall Street Journal]

Nvidia's H20 chip orders surge as Chinese firms adopt DeepSeek's AI models - Chinese companies, including tech giants and smaller firms, are increasing orders for Nvidia's H20 AI chips due to the rising demand for DeepSeek's cost-effective AI models. [Reuters]

Anthropic launches world's first 'hybrid reasoning' AI model - The new model, Claude 3.7, combines instinctive output with in-depth reasoning, allowing users to adjust the degree of reasoning for complex problem-solving. [WIRED]

Meta discusses $200 billion AI data center project - Meta Platforms is reportedly in talks to build a new data center campus dedicated to AI projects, potentially costing over $200 billion, with possible locations including Louisiana, Wyoming, or Texas. [Reuters]

OpenAI expands deep research access to all paying ChatGPT users - OpenAI has made its deep research features available to all subscribers, enhancing the capabilities of ChatGPT for complex inquiries. [Engadget]

Anthropic's Claude AI plays Pokémon on Twitch - Demonstrating advanced AI capabilities, Anthropic's Claude AI engages in playing Pokémon live on Twitch, showcasing its interactive learning potential. [TechCrunch]

Groks' uncensored AI voice mode explores sensitive topics - The new voice mode allows users to discuss sensitive subjects, including sex therapy and conspiracies, raising discussions about AI's role in handling delicate conversations. [Ars Technica]

Y Combinator deletes posts after startup's demo goes viral - Following a startup's viral demo, Y Combinator removed related posts, sparking conversations about content management and publicity within startup communities. [TechCrunch]

Activision confirms AI use in 'Call of Duty' development - The gaming company acknowledges the use of AI tools in creating in-game assets for the popular franchise, validating long-held player suspicions. [Dexerto]

⚖ Legal

Education tech firm sues Google over AI search summaries - Chegg files a federal antitrust lawsuit against Google, alleging that AI-generated summaries in search results have significantly reduced traffic to its website, impacting revenue and employment. [Ars Technica]

Musicians release silent album to protest UK's AI copyright changes - Over 1,000 musicians, including Kate Bush and Cat Stevens, release a silent album titled "Is This What We Want?" to oppose proposed UK copyright law changes that would allow AI developers to use artists' works without compensation. [Reuters]

🎱 Random

Disney employee's AI tool download leads to hack - A Disney worker's download of an AI tool resulted in a cyberattack, highlighting the security risks associated with unauthorized software use. [The Wall Street Journal]

Apple investors reject call to end DEI efforts at annual meeting - Shareholders vote against a proposal to terminate diversity, equity, and inclusion initiatives, reaffirming the company's commitment to these programs. [Bloomberg]

Philip Moyer discusses Vimeo's AI strategy and competition with YouTube - In an interview, Vimeo's CEO outlines the company's AI-driven approach to support creators and compete with platforms like YouTube. [The Verge]

🔌 Plug-Into-This

Alibaba has announced the open-source release of its AI model, Wan 2.1, capable of generating realistic videos and images. This strategic move aims to enhance adoption and competitiveness in the AI sector.

  • Model Variants: Wan 2.1 includes four versions—T2V-1.3B, T2V-14B, I2V-14B-720P, and I2V-14B-480P—designed to process varying levels of input complexity, with "14B" models handling 14 billion parameters for more precise outputs.

  • Accessibility: These models are available globally on Alibaba Cloud's ModelScope and HuggingFace platforms, supporting academic, research, and commercial applications.

  • Performance Benchmark: Wan 2.1 has achieved top rankings on VBench, a leaderboard for video generative models, particularly excelling in multi-object interaction capabilities.

  • Investment Commitment: The company has pledged to invest at least 380 billion yuan ($52 billion) over the next three years to strengthen its cloud computing and AI infrastructure.

🎥 This release positions Alibaba alongside other eastern innovators like DeepSeek with the introduction of advanced models into the open-source realm — early results look impressive.

Anthropic has unveiled Claude 3.7 Sonnet, an advanced language model featuring an "extended thinking mode" that enhances its reasoning capabilities. This release aims to provide users with more comprehensive and contextually aware AI-generated content.

  • Extended Thinking Mode: This feature allows Claude 3.7 Sonnet to allocate up to 128,000 tokens for internal reasoning, enabling deeper analysis and more nuanced responses.

  • Enhanced Output Capacity: The model can generate outputs up to 128,000 tokens in length, significantly surpassing previous versions and facilitating more detailed content generation.

  • Performance Evaluation: Through his llm-anthropic plugin, Willison tested the model's capabilities, noting its proficiency in generating complex outputs, such as detailed speeches, and observing that longer outputs proportionally increase processing time and cost.

  • Comparison with Other Models: Claude 3.7 Sonnet's extended thinking mode positions it alongside models like OpenAI's o1 and o3, offering users enhanced reasoning options.

🧠 The introduction of extended thinking mode in Claude 3.7 Sonnet reflects a broader trend in AI development, emphasizing the importance of internal reasoning processes to produce more accurate and contextually relevant outputs.

Google has introduced a free version of Gemini Code Assist, an AI-powered coding tool designed to support individual developers, including students, freelancers, and hobbyists. This move expands access to advanced AI-driven coding assistance without subscription fees, positioning Google as a strong competitor in the AI coding assistant space.

  • Generous Usage Limits: Users can access up to 180,000 code completions per month, far exceeding the limits of competing services like GitHub Copilot, which offers only 2,000 completions per month.

  • Powered by Gemini 2.0: Built on Google's Gemini 2.0 AI model, the tool supports code generation and completion across multiple programming languages through an intuitive chatbot interface.

  • IDE Compatibility: Gemini Code Assist integrates seamlessly with major development environments such as Visual Studio Code, GitHub, and JetBrains, ensuring smooth adoption by developers.

  • Advanced Features: While the free version provides robust functionality, more advanced tools—including productivity metrics and deeper integrations with Google Cloud services—remain exclusive to the paid Standard and Enterprise tiers.

  • AI-Powered Code Review: The tool includes intelligent code review capabilities for both public and private repositories, helping developers streamline their workflows and improve code quality.

🧐 While Google's decision to offer Gemini Code Assist for free appears generous, it may also signal challenges in driving adoption. By providing advanced features at no cost, Google could be attempting to lure developers into its ecosystem after struggling to compete with entrenched tools like GitHub Copilot. If uptake remains low, this strategy might hint at deeper issues in convincing users to switch.

 🆕 Updates

📽️ Daily Demo

🗣️ Discourse