Anthropic’s Claude can now automatically control your computer!
Plus tons of updates from OpenAI, Meta, Runway, Perplexity, Stability, and more!
Today’s Highlights:
📰 News: Anthropic’s Claude can now automatically control your computer! + tons of updates from OpenAI, Meta, Runway, Perplexity, Stability, and more!
💰 Funding: Infinitus Systems raises $50M+ to automate healthcare phone calls
⚡️ Top News Stories:
1. Anthropic Introduces ‘Computer Use’ for Claude, a tool that can autonomously navigate computer interfaces, performing tasks like viewing screens, typing, and executing commands across applications and websites.
They also announced some updates to models and tools:
Sonnet 3.5 Improvements: The upgraded Sonnet 3.5 offers major enhancements in coding and tool use, outperforming previous models on benchmarks such as SWE-bench.
Haiku 3.5 Release: Haiku 3.5 is more affordable and faster, matching the performance of larger models like Claude 3 Opus.
Analysis Tool: Enables users to run JavaScript for real-time data analysis, supporting tasks like processing CSV files, performing calculations, and generating insights across sectors such as marketing and finance.
2. AI startup Genmo has introduced Mochi 1, an open-source text-to-video generation model built on a 10Bn parameter architecture, claiming state-of-the-art motion quality and prompt adherence.
Mochi 1 focuses on smooth motion and accurate prompt adherence, generating 480p videos at 30fps for up to 5.4 seconds, with HD video and more advanced features expected later this year
Mochi 1 is freely available under the Apache 2.0 license, allowing developers to use it for research and commercial applications.
Genmo raised $28.4M in Series A funding to further develop AI tools, aiming to rival closed models like Runway and Pika.
3. President Biden mandates AI guardrails for intelligence and national security agencies, focusing on human oversight.
President Biden signed a national security memorandum establishing AI guardrails for agencies, including prohibitions on autonomous nuclear launches and limiting AI’s role in targeting, asylum, and tracking decisions.
The memorandum also mandates protections for AI technology to prevent espionage, using the AI Safety Institute to ensure AI safety against terrorist misuse.
Annual reports will assess AI's risks in nuclear and biosecurity contexts, with an aim to engage China on limiting high-risk AI applications.
4. Runway has unveiled Act-One, a tool that enables creators to map real human facial expressions onto AI-generated characters using just a smartphone video and a reference image, capturing subtle micro-expressions, eye movements, and facial nuances for impressive realism without the need for specialized equipment.
5. Perplexity has introduced Internal Knowledge Search, enabling Pro and Enterprise Pro users to search both internal files and public web content for faster, integrated research, with use cases like financial services firms conducting due diligence, sales teams streamlining the RFP process, and HR teams quickly answering employee questions using internal knowledge bases.
6. Stability AI introduces Stable Diffusion 3.5, featuring models like Large, Large Turbo, and Medium, emphasizing customizability, speed, and efficiency on consumer hardware.
7. Asana has introduced AI Studio, a no-code tool for building AI-driven workflows with AI agents that streamline task management by acting as teammates and handling routine tasks such as intake, planning, and reporting.
8. Inflection AI has launched Agentic Workflows for enterprise users, allowing AI agents to automate tasks using accurate, up-to-date information, with a partnership with UiPath to ensure actions align with company policies for improved workflow reliability, while focusing on IQ, EQ, and AQ (action quotient) to build high-performance, actionable AI systems.
9. Ideogram Launches Canvas, a new AI-powered workspace that allows users to generate, edit, and combine images on an infinite digital board.
10. Google has open-sourced SynthID, a watermarking tool for AI-generated text, that embeds detectable watermarks without affecting output quality, helping developers identify AI-created content and addressing concerns over misinformation and malicious use of AI.
11. Meta has released smaller, compressed versions of its Llama 3.2 AI models, designed to run on smartphones and tablets, allowing AI to operate without large data centers.
12. OpenAI has introduced sCM (simplified continuous-time consistency models), a breakthrough that generates high-quality AI samples with just two sampling steps, offering a 50x speedup over traditional diffusion models, drastically reducing computational costs while maintaining comparable quality.
13. OpenAI plans to launch Orion, a powerful AI model rumored to be 100x stronger than GPT-4, in December, initially for select partner companies.
14. Former OpenAI researcher Suchir Balaji has criticized the company for using copyrighted data to train AI models like GPT-4, arguing that it violates laws, harms content creators, and does not meet "fair use" standards. He also highlights the risks of AI "hallucinations" and false information, calling for regulatory intervention to address AI's rapid evolution.
15. Miles Brundage, OpenAI’s former senior adviser for AGI readiness, has left the company, warning that neither OpenAI nor the world is prepared for AGI. His departure is the latest to highlight internal tensions between OpenAI’s safety-focused mission and its growing commercial ambitions. Brundage also urged OpenAI employees to voice concerns and cautioned against groupthink.
16. Google Photos will now be labeling AI-edited images in the metadata, such as those modified with Magic Editor and Magic Eraser, though visible watermarks on the images themselves will remain absent.
17. Nvidia is expanding in India through partnerships with Reliance, Tech Mahindra, Tata Communications, and Yotta Data Services to build AI infrastructure, while also training thousands of developers with Wipro and TCS as India pivots from software exports to AI.
💰 Top Funding News:
1. Infinitus Systems, which automates healthcare operations, specifically manual phone calls, using AI-driven systems with strict guardrails to ensure compliance and accuracy, raised a $51.5M Series C, led by Andreessen Horowitz, w/ Memorial Hermann Health System, Kleiner Perkins, Coatue, and GV (Google Ventures).
2. Socket, which uses AI-powered tools to proactively monitor open-source software for malicious behaviors such as backdoors, typo-squatting, and obfuscated code, raised a $40M Series B, led by Abstract Ventures, w/ Elad Gil, Andreessen Horowitz, and several angel investors.
3. Tennr, which uses ML and document scanning tech to automate the extraction of referral information from various formats, raised a $37M Series B, led by Lightspeed Ventures, with participation from Andreessen Horowitz and Foundation Capital.
4. Fixify, which uses AI to support IT help desk teams by automating the resolution of common issues such as software problems and password resets, raised a $25M Series A, co-led by Costanoa Ventures, Decibel Partners, and Paladin Capital Group, with participation from Scale Venture Partners.