AI Weekly Update - October 18

Adobe previews future AI tools, ChatGPT gets a Windows app

what to know for now

💻 OpenAI launches ChatGPT Windows app. OpenAI releases a dedicated ChatGPT app for Windows, offering access to GPT-4o, o1-preview, and associated features. The app supports file uploads but lacks voice features, with more functionality coming later. Read more

🦙 NVIDIA releases powerful Llama-3.1-Nemotron model. NVIDIA’s open-source Llama-3.1-Nemotron-70B-Instruct surpasses OpenAI’s GPT-4o and Claude 3.5 in benchmarks. The model uses advanced reward modeling to refine responses, making it highly accurate in instruction-based tasks. Read more

🎙️ NotebookLM enhances AI podcasting. Google’s NotebookLM adds new customization features for creating personalized AI-generated podcasts. Users can now guide topics and adjust expertise levels while querying sources in real-time. NotebookLM Business introduces a paid version with premium features. Read more

📱 Claude AI expands mobile usability. Anthropic launched updates for Claude AI, introducing an iPad app and enhanced user customization features. These updates include the ability to search past chats and personalize instructions, increasing flexibility across various devices for improved productivity. Read more

📰 NYT demands Perplexity stop using its content. The New York Times sent a cease-and-desist letter to Perplexity AI, accusing it of unauthorized content use. Perplexity claims it's indexing facts, not scraping for model training, and plans to respond by October 30th. Read more

🧪 AI Research of the Week

Movie Gen: A Cast of Media Foundation Models
From Meta

Jake’s Take: This paper introduces Movie Gen, a suite of large media generation models developed by Meta, focused on high-quality video, image, and audio generation. These models support a range of tasks, including text-to-video synthesis, personalized video creation based on user images, video editing, and synchronized audio generation. The largest model, a 30B-parameter transformer, can generate 16-second HD videos at 16 frames-per-second with various aspect ratios, significantly advancing the field of media generation. Innovations span the architecture, data curation, and inference processes to achieve state-of-the-art results, outperforming existing commercial systems.

The industry should to be prepared for a future where personalized, high-definition media content can be generated on demand, reshaping everything from short form entertainment to advertising.

what to know for later

🎨 Adobe debuts cutting-edge creative tools. Adobe MAX introduces new Sneaks projects transforming workflows for photos, videos, audio, and 3D design. Innovations include AI-driven tools like Project Perfect Blend for realistic photo edits and Project Super Sonic for intuitive sound design. Read more

🐝 OpenAI launches Swarm for agentic AI. OpenAI’s Swarm product enables lightweight multi-agent orchestration, allowing AI agents to autonomously handle complex tasks. This agentic AI framework focuses on collaboration between AI agents, bringing more automation to real-world applications. Read more

🔄 Google restructures search leadership. Google replaces search chief Prabhakar Raghavan amid Gemini AI shifts. Raghavan moves to a technologist role while the Gemini chatbot moves to DeepMind. This restructuring may influence the future of search quality and AI integration. Read more

🎥 Adobe launches Firefly AI video model. Adobe introduces Firefly Video Model for Premiere Pro, enabling users to extend footage and generate videos from text or images. Currently limited to short clips with basic resolution, the tool remains in beta. Read more

🔬 AI uncovers new physics laws. Archetype AI's Newton model learns complex physics from raw sensor data without any pre-programming. It can generalize across diverse phenomena like mechanical oscillations and thermodynamics, outperforming specialized models in real-world scenarios such as power grid predictions. Read more