AI News: Anthropic Went Crazy This Week!

This week has seen a significant surge of news and updates in the AI world, with companies like Anthropic and Google leading the charge with new features and model releases. Anthropic, in particular, has been consistently shipping new functionalities, with reports indicating 74 releases in just 52 days. Recent additions include cloud tasks, projects, co-work domain selection, effort and skills tracking, various cloud code updates, computer use capabilities, scheduled cloud code execution, cloud code channels for teams and enterprise, and the ability to interact with tools like Figma, Amplitude dashboards, and canvas slides from a mobile phone using Claude. A particularly impactful feature for Anthropic is the "computer use" capability for Claude. Launched on the 23rd, this allows Claude to control a user's computer, including mouse and keyboard actions. When combined with the "dispatch" feature, which enables control of Claude Code and Claude Co-work from a phone, users can remotely instruct their computers to perform tasks. While this feature requires a paid plan and needs to be enabled in settings, it offers the potential for hands-free computer operation. However, the current implementation is noted to be quite slow, with initial attempts at tasks like opening applications taking considerable time. The speaker demonstrated opening Da Vinci Resolve, which took approximately five minutes, highlighting that while functional, it's not yet efficient for tasks that can be done quickly manually. The primary benefit of this slow but capable feature appears to be for remote task execution where time is less critical than task completion. Claude Co-work has also received a significant update with the introduction of a "projects" ability, allowing for better organization of work within Claude. Users can create projects and assign custom instructions to guide Claude's behavior within that project. For instance, a project could be set to have Claude respond in a specific persona, like a "gym bro," and incorporate files for added context. This functionality is similar to project management features found in other AI platforms. For AI users involved in coding, "auto mode" for Claude Code is a welcome addition. Previously, Claude Code would frequently pause to ask for permission for actions like browsing the web or running terminal commands, disrupting workflow. Auto mode now handles many of these more benign commands without requiring explicit permission, which is expected to be a significant improvement for Claude Code users. Beyond Anthropic, GenSpark has emerged as a notable AI workspace offering an all-in-one platform for presentations, data analysis, and report writing. A key draw for GenSpark is its offer of unlimited AI chat and image generation with its paid plan until December 31st, 2026, for $20 per month. This is particularly attractive as individual access to similar models can cost around $20 per month each. GenSpark aims to streamline content creation, from presentations and spreadsheets to websites, images, and videos, all within a single platform. Google has also been busy with releases, notably Gemini 3.1 Flash Live. This new model is integrated across various Google products, including their API, enterprise solutions, search, and Gemini Live. Gemini 3.1 Flash Live supports interactive, multimodal conversations, allowing users to share their webcam and screen. The speaker demonstrated its ability to identify objects on a webcam feed and understand content on a shared screen, showcasing its potential for teaching and guidance, particularly within applications like OBS Studio for streaming. This feature is also being integrated into Google Search, enabling interactive AI-powered conversations. Google also showcased a novel browser built with Gemini 3.1 Flash that generates new web pages in real-time based on user input, though it currently lacks memory and is primarily a demonstration of capability. Recognizing the need for user migration, Google has also made it easy for users to transfer their chat history, memories, and preferences from other AI platforms like ChatGPT and Claude to Gemini. Furthermore, Google released Lia 3 Pro, a music generation model capable of producing longer tracks (up to three minutes) and allowing for prompts that specify song structures like intros, verses, and choruses. This is an expansion from its previous 30-second song generation limit and is being integrated into various Google products. The music generation space also saw an update from Souno with version 5.5, which allows users to train their own voices into the model. While the speaker humorously demonstrated the potential for less-than-ideal singing with a cloned voice, it highlights the growing personalization capabilities in AI music generation. In text-to-speech, Small.ai has introduced a new model designed for conversational voice agents, aiming to compete with services like Eleven Labs. This model is tuned to sound more natural, mimicking thinking and listening pauses, and is designed to keep users engaged. It also offers voice cloning capabilities. Mistral released an open-weight text-to-speech model that can be run locally. This model is presented as a strong competitor to Eleven Labs, with a blind test suggesting it may even outperform it in some aspects. It is also capable of capturing nuances like accents and inflections, and can adapt to custom voices with very short reference samples. For AI art enthusiasts, Love Art AI has launched a "move object" feature. This allows users to upload an image, highlight an object within it, and reposition it to generate a modified image. This can then be used with video generation tools to create animations where the object moves. In a significant development, OpenAI has decided to cease focusing on video generation capabilities, including the Sora app, video generator, and API. This decision stems from a need to concentrate on their core chat and coding models, which are seen as more profitable and aligned with user demand. This move has led to the cancellation of a deal with Disney, which was contingent on access to OpenAI's video technology. Additionally, plans for an "adult mode" in ChatGPT have also reportedly been scrapped as part of this strategic shift. The speaker notes a parallel between these abandoned projects and Elon Musk's current focus at XAI. Concerns have also been raised about the effectiveness of OpenAI's advertising platform. Early advertisers have reported a low-tech process and a lack of measurable business outcomes from their campaigns on ChatGPT. The speaker speculates that the placement of ads within a conversational AI interface may not be conducive to user engagement. Despite these challenges, OpenAI is expanding its e-commerce integrations within ChatGPT, allowing for richer product discovery and easier listing for businesses, suggesting a potential future for featured product placements as an advertising model. OpenAI has also integrated plugins into its CodeX app, enabling the use of tools like Slack, Figma, and Notion within their IDE. In legal news, Anthropic has achieved a preliminary victory against a designation as a supply chain risk by the Department of War. A US federal judge halted one of the designations, citing free speech concerns. However, another designation remains pending, meaning Anthropic still faces legal challenges. A leaked document from Anthropic suggests the development of a new generation of powerful models called "Claude Mythos," which are described as larger and more intelligent than their current Opus models. These models are expected to excel in areas like software coding, academic reasoning, and cybersecurity. Anthropic has expressed caution regarding the potential near-term risks, particularly in cybersecurity, and the high cost of development and usage. CapCut has released Drama's Seed Dance 2.0, a video model that has been met with significant attention, though its availability in the US and Europe is currently restricted due to trademark and likeness concerns. Wikipedia has banned AI-generated articles, allowing AI only for basic copy editing and translation. This move is seen as a measure to prevent model collapse, given that many LLMs were trained on Wikipedia data. Finally, in robotics news, the Figure03 humanoid robot made a historic visit to the White House, marking the first time a humanoid robot has been present there.