Today:

• OpenAI Revenue Doubles This Year

• Former Meta Engineers Launch Jace

• Gemini Integration in Google Messages

• Luma's Dream Machine Faces Surge

• Stability AI Launches Advanced Model

• Databricks Unveils Mosaic AI Updates

• Google AI Enhances Health Insights

• Shareholders Sue Musk for xAI Launch

Newsletter advertising at scale, finally.

The beehiiv Ad Network delivers over a billion  impressions a month across thousands of the world's top newsletters, all paid on performance. 

Our managed service enables you to unlock newsletters as a growth channel without the added time. 

Learn more now.

OPENAI

• OpenAI, backed by Microsoft, has doubled its annualized revenue to $3.4 billion, with $3.2 billion from subscriptions and $200 million from Microsoft sales. This marks significant growth from $1.6 billion last year. ChatGPT remains a popular app, and Apple will integrate ChatGPT-4o into Siri.

• Apple announced a partnership with OpenAI to integrate ChatGPT into iPhones, iPads, and Macs, revealed at the Apple Worldwide Developers Conference. Financial terms remain undisclosed, but sources indicate the collaboration won’t initially generate significant revenue for either company.

• Oracle, Microsoft, and OpenAI are partnering to expand the Microsoft Azure AI platform to Oracle Cloud Infrastructure (OCI), enhancing capacity for OpenAI's services. This collaboration leverages OCI's AI infrastructure, enabling OpenAI to scale efficiently.

• Brazil's government has hired OpenAI to use artificial intelligence to screen and analyze lawsuits, aiming to reduce costly court losses. The AI will help flag actions needed before final decisions. Microsoft will provide these AI services through its Azure cloud platform.

• Elon Musk has withdrawn his lawsuit against OpenAI and its co-founders, Sam Altman and Greg Brockman, in California state court. The case, originally filed in February, alleged breach of contract and fiduciary duty. Experts questioned the legal foundation of the suit. Musk's decision follows public criticism of OpenAI's partnership with Apple.

• Mira Murati, OpenAI's CTO, defended the company's new partnership with Apple against Elon Musk's criticism, calling the integration of ChatGPT into Apple devices "creepy spyware." Murati emphasized OpenAI's commitment to user privacy and transparency, clarifying that Apple won't share user data with OpenAI. Musk had publicly denounced the alliance and threatened to ban Apple devices at his companies.

Former Meta engineers launch Jace, an AI agent that works independently

Former Meta engineers Fryderyk Wiatrowski and Peter Albert have launched Jace, an AI agent developed by their startup Zeta Labs. Jace, powered by a large language model (LLM), can independently perform browser tasks without user intervention. The startup raised $2.9 million in pre-seed funding to enhance Jace's capabilities. 

Jace can handle tasks from booking hotels to managing inventory, aiming to automate repetitive tasks for consumers and businesses. Zeta Labs is refining Jace for general release, planning a subscription model at $45/month, and targeting sectors like recruiting, ecommerce, and marketing.

VENTUREBEAT

Google Messages is about to put Gemini front and center (APK teardown)

Google is streamlining access to its Gemini chatbot in the Messages app. A new floating action button (fab) will soon allow users to start a conversation with Gemini directly, bypassing the current multi-step process. 

This update, found in an APK teardown of version messages.android_20240610_01_RC00, places the Gemini logo above the existing Start chat button for quicker access. The smaller size of the new button is noted, though it's unclear if this is intentional. This change aims to simplify interactions and potentially increase user engagement with the AI chatbot. The feature is expected to be available to beta users soon.

ANDROID AUTHORITY

‘We don’t need Sora anymore’: Luma’s new AI video generator Dream Machine slammed with traffic after debut

Luma AI launched the public beta of its AI video generator, Dream Machine, backed by Andreessen Horowitz. The tool, generating up to 120 frames in 2 minutes, faced high demand, causing long queues. Luma's team is working to increase capacity, promising normal processing times of 2-3 minutes. The tool's high-quality output impressed early testers, despite some accuracy issues. 

Luma, previously known for its text-to-3D asset generator Genie 1.0, raised over $70 million, including $43 million in Series B funding. Dream Machine is competing with OpenAI's Sora and other AI video models like Runway and Pika.

VENTUREBEAT

Announcing the Open Release of Stable Diffusion 3 Medium, Our Most Sophisticated Image Generation Model to Date 

Stability AI has launched Stable Diffusion 3 Medium, their most advanced text-to-image model. This 2 billion parameter model excels in photorealism, understanding complex prompts, and efficient use of resources, making it suitable for consumer and enterprise GPUs. 

Released under a non-commercial license with a low-cost Creator License for commercial use, it can be tried via API or on Stable Assistant and Stable Artisan. Collaborations with NVIDIA and AMD enhance its performance. Stability AI emphasizes safety and responsible AI use, with plans for continuous improvement based on user feedback.

STABILITY AI

Databricks expands Mosaic AI to help enterprises build with LLMs

Databricks has expanded its Mosaic AI platform, acquired last year for $1.3 billion, with new features to aid enterprises in building large language models (LLMs). Announced at the Data + AI Summit, the new tools include Mosaic AI Agent Framework, Evaluation, Tools Catalog, Model Training, and Gateway. 

These tools aim to improve model reliability, cost-efficiency, and data privacy. The updates integrate modular systems for better control and performance, enabling companies to build, fine-tune, and manage AI models effectively. This shift reflects a growing demand for sophisticated, open-source AI solutions tailored to specific enterprise needs.

TECHCRUNCH

Advancing personal health and wellness insights with AI

Google Research has introduced a novel large language model (LLM) to enhance personal health insights using data from mobile and wearable devices. This model, fine-tuned from Gemini, offers personalized health recommendations by analyzing physiological data like heart rate and sleep patterns. Google has curated three benchmark datasets to evaluate the model's effectiveness in providing health insights, expert-level knowledge, and predicting self-reported sleep outcomes. 

Additionally, a new framework enables the model to analyze wearable data and generate tailored health recommendations. This research aims to improve personal health management by offering precise, actionable insights.

GOOGLE

Tesla investors sue Elon Musk for launching a rival AI company

Several Tesla shareholders have filed a lawsuit against Elon Musk and Tesla's board, accusing them of diverting resources to Musk's new AI company, xAI. The shareholders claim that Musk breached his fiduciary duty by launching xAI and redirecting talent and resources from Tesla to the startup. 

The lawsuit, filed in Delaware, alleges that Musk raised billions for xAI while leveraging Tesla's AI-related data. Additionally, the plaintiffs cite Musk's diversion of Nvidia AI chips intended for Tesla to xAI and his push for a larger stake in Tesla to solidify its AI ambitions. This lawsuit follows another alleging Musk's insider trading with Tesla stock.

THE VERGE

🧠RESEARCH

NaRCan: Natural Refined Canonical Image with Integration of Diffusion Prior for Video Editing

NaRCan is a video editing framework that enhances video quality by integrating deformation fields and diffusion prior. It uses homography for global motion and MLPs for local deformations, ensuring high-quality images. The model accelerates training by 14 times, outperforming current methods in various video editing tasks.

An Image is Worth 32 Tokens for Reconstruction and Generation

TiTok is a novel Transformer-based tokenizer that converts images into compact 1D latent sequences, significantly reducing the number of tokens needed for high-resolution image generation. This method outperforms traditional 2D tokenization techniques, providing faster and more efficient image synthesis. TiTok achieves state-of-the-art performance, notably improving generation speed and quality.

Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation

LlamaGen introduces a family of image generation models that use the "next-token prediction" method from language models. It features a high-quality image tokenizer and scalable models that outperform popular diffusion models in image quality. LlamaGen also achieves significant speed improvements in image generation, enhancing both visual quality and text alignment.

Vript: A Video Is Worth Thousands of Words

Vript is a comprehensive video-text dataset featuring 12,000 high-resolution videos with detailed, script-like captions for over 420,000 clips. Each caption averages 145 words, documenting both content and camera operations. This dataset enhances video understanding and generation. The Vriptor model, trained on Vript, excels in video captioning, rivaling GPT-4V. The benchmark, Vript-Hard, introduces challenging tasks for video understanding.

McEval: Massively Multilingual Code Evaluation

McEval introduces a multilingual code benchmark covering 40 programming languages with 16,000 test samples, enhancing code understanding, completion, and generation tasks. It includes McEval-Instruct, a multilingual instruction corpus, and mCoder, a model trained on it. Despite progress, open-source models still lag behind closed-source models like GPT. The benchmark and resources are available at the McEval website.

🛠️TOP TOOLS

Luma Dream Machine -  Creates high-quality, realistic videos from text and images, utilizing cutting-edge technology.

Udio - Generate AI music instantly.

Recall - Summarize any online content and save it to your knowledge base where it’s automatically organized and interlinked for easy rediscovery

Capcut - Free all-in-one video editor for everyone to create anything anywhere

ElevenLabs Generative Voice AI - Convert text to speech online for free with our AI voice generator.