Anthropic’s AI Agent

Anthropic’s Claude 3.5 Sonnet introduces AI that may interact with a pc very similar to a human. Now in beta, this model allows developers to program the AI to view screens, move cursors, click buttons, and input text using an API. While still experimental and susceptible to errors, Claude 3.5 Sonnet is already getting used by firms like Asana, Canva, and Replit to automate tasks and test software.

Claude “learns” by processing screenshots and counting pixels to determine where to point the cursor, but more advanced moves like dragging windows still elude it. Despite some missteps, it’s already outperforming other models in tests, slowly closing the gap between AI and human-level interaction.

Available on platforms like Amazon Bedrock and Google Cloud’s Vertex AI, Claude is a work-in-progress, with Anthropic fine-tuning its abilities through developer feedback, aiming for smoother, more reliable performance in computer tasks down the road. Read more.

NetVRk has launched the alpha version of its AI-driven metaverse, built on Unreal Engine 5. This platform incorporates a Language Model (LLM) to reinforce interactions, allowing players to have interaction with emotionally intelligent non-player characters (NPCs) that exhibit real-time memory and responsiveness.

Users can customize and trade these NPCs as non-fungible tokens (NFTs) while earning rewards. With a team of about 14 members and $10 million raised, NetVRk’s vision combines AI, user-generated content, and blockchain functionalities to redefine virtual engagement and interaction. Read more.

Ideogram, a Canadian AI startup, has launched “Canvas,” an infinite workspace allowing users to control and mix generated images. This feature enables resizing, reordering, and uploading personal visuals.

Known for its precise text embedding in images, Ideogram also introduced two latest tools: Magic Fillwhich edits specific areas using text prompts, and Extendwhich expands images while maintaining style consistency. Available through various subscription plans. Read more.

Asana has launched AI Studio, a no-code tool that empowers teams to create AI agents designed to optimize workflows. Introduced on the Work Innovation Summit, this feature extends Asana’s existing rules engine, allowing users to customize AI agents for various project phases, from intake to reporting. Teams can upload context-specific data, tailoring AI-generated content and task updates. Positioned as virtual teammates, these agents help with tasks like drafting materials and tracking efficiency, aiming to alleviate the 53% of worker time spent on low-value tasks. Early access is on the market for Enterprise customers, with advanced tiers coming soon. Read more.

Runway AI Inc. has introduced Act-One, a tool that simplifies facial animation in AI-generated video. Users can record their facial expressions using a smartphone camera, mapping them onto digital characters to reinforce animation quality. While available to all Runway users, advanced features of the Gen-3 Alpha model require credits.

Act-One allows for improved facial accuracy, enabling creators to copy diverse expressions on avatars, making it accessible for animators, game developers, and indie filmmakers without the necessity for costly equipment. Built-in safeguards promote ethical use, allowing users to explore storytelling in a responsible manner. Read more.

• Stability claims its newest Stable Diffusion models generate more ‘diverse’ images

• Elon Musk’s xAI API launches, letting developers construct atop Grok

  Gartner: 2025 will see the rise of AI agents (and other top trends)

  Microsoft releases autonomous Copilot agents from closed beta

• AI video startup Genmo launches Mochi 1, an open source rival to Runway, Kling, and others

• Robots learn to perceive objects using acoustic vibrations

• ByteDance intern fired for planting malicious code in AI models

• AI Uncovers DNA Variants Linked to Psychiatric Disorders

• ‘Blade Runner 2049’ Producers Sue Elon Musk, Tesla and Warner Bros. Discovery, Alleging Copyright Infringement

• News Corp sues Perplexity for ripping off WSJ and Latest York Post

• Daze, a creative, AI-powered messaging app for Gen Z, is blowing up prelaunch

• Google DeepMind and MIT’s latest “Fluid” model outperforms diffusion models in image generation

• Cohere adds vision to its RAG search capabilities

• OpenAI Hires Aaron Chatterji to Be Its Chief Economist

• Agent S AI system automates complex computer tasks

• Greater than 11,000 creatives condemn unauthorized use of content for AI development

• Microsoft and OpenAI are giving news outlets $10 million to make use of AI tools

• ChatGPT Advanced Voice Mode now available within the EU