Today:

• Character.AI Founders Join Google

• OpenAI Grants US Early Model Access

• Nvidia Delays Blackwell AI Chips

• GitHub Launches AI Development Platform

• Google Pulls Gemini AI Ad

• UK Cuts £1.3bn Tech Funding

• Cloud Infrastructure Hits $80B

• Stable Diffusion Team Unveils FLUX.1

• OpenAI Enters New VC Phase

Google hires founders of chatbot start-up Character.AI

Google has hired the founders of Character.AI, Noam Shazeer and Daniel De Freitas, and licensed the start-up’s language model technology. Shazeer will join Google’s DeepMind, while most of Character.AI’s team remains. The deal, reflecting Big Tech’s interest in AI start-ups, allows Character.AI to leverage Google’s resources for further growth. 

Character.AI, known for creating AI avatars, has millions of users and competes with similar tools from companies like Meta. This move helps Google avoid antitrust scrutiny and continues the trend of tech giants partnering with innovative AI firms.

FINANCIAL TIMES

OpenAI vows to provide the US government early access to its next AI model

OpenAI will grant the US AI Safety Institute early access to its next AI model as part of its commitment to safety. This collaboration aims to advance AI evaluation science. The Institute, established by the National Institute of Standards and Technology (NIST), creates guidelines and standards for AI safety. 

Despite concerns that OpenAI prioritizes powerful AI development over safety, CEO Sam Altman confirmed the company’s dedication to security. Internal issues, including disbanding the Superalignment team and leadership changes, have raised further safety concerns. OpenAI’s new safety group is now led by board members, including Altman.

ENGADGET

Nvidia reportedly delays its next AI chip due to a design flaw

Nvidia's "Blackwell" B200 AI chips will be delayed by at least three months due to a design flaw found late in production, reports The Information. This delay affects Microsoft and another cloud provider, with mass production now expected in the first quarter of next year. 

These chips follow the popular H100 chips and are crucial for AI cloud infrastructure. Nvidia is conducting new test runs with Taiwan Semiconductor Manufacturing Company. The delay comes after Nvidia announced that Blackwell-based products would be available in 2024, aiming for an annual release of new AI chips amidst rising competition.

THE VERGE

Introducing GitHub Models: A new generation of AI engineers building on GitHub

GitHub has launched GitHub Models, empowering its 100 million users to become AI engineers. This platform provides access to various large and small language models, allowing developers to test, compare, and deploy AI applications directly on GitHub. Users can experiment with models like Llama 3.1 and GPT-4o in an interactive playground, then seamlessly integrate them into their projects using Codespaces and VS Code. 

Once ready for production, developers can deploy on Azure AI with enterprise-grade security and global availability. GitHub Models aims to democratize AI, enabling a vast community of developers to innovate and build AI-driven applications.

GITHUB

Google Pulls Controversial ‘Dear Sydney’ Gemini AI Ad From Olympics Coverage—What To Know

Google has withdrawn its “Dear Sydney” Gemini AI ad from Olympics broadcasts after it faced backlash for promoting artificial intelligence over human interaction. The ad featured a father using AI to write a letter to American hurdler Sydney McLaughlin-Levrone on behalf of his daughter. 

Critics found the ad disturbing, arguing it undermined the personal nature of writing a fan letter. Despite Google’s defense that the ad celebrated Team USA, they decided to phase it out. This controversy adds to ongoing debates about AI’s role in everyday life, with concerns about AI replacing meaningful human activities.

FORBES

UK shelves £1.3bn of funding for technology and AI projects

The UK Labour government has canceled £1.3 billion in funding for technology and AI projects, initially promised by the Conservatives. This decision jeopardizes the UK's first next-generation supercomputer planned for the University of Edinburgh. The £800 million exascale supercomputer project and £500 million for the AI Research Resource are affected. 

The government cites the need to address “unfunded commitments” and restore economic stability. Despite this, £300 million for the AI Research Resource has already been allocated. The decision has sparked criticism from the tech sector, which warns it could hinder the UK's technological advancements and competitiveness.

THE GUARDIAN

Cloud infrastructure revenue approached $80 billion this quarter

The cloud infrastructure market surged to nearly $80 billion this quarter, driven by AI advancements, marking a 22% increase from last year, according to Synergy Research. This is the third consecutive quarter of 20%+ year-over-year growth. 

Microsoft's Intelligent Cloud reported $28.52 billion, slightly missing analyst expectations but still growing 30% year-over-year. Amazon saw a 19% revenue increase to $26.3 billion, while Google Cloud surpassed $10 billion, growing 29%. The Big 3—Amazon, Microsoft, and Google—held 32%, 23%, and 12% of the market share, respectively. Oracle moved up to 3%, with the market expected to double in four years.

TECHCRUNCH

Stable Diffusion creators launch Black Forest Labs, secure $31M for FLUX.1 AI image generator

Black Forest Labs, founded by the creators of Stable Diffusion, launched its FLUX.1 text-to-image model suite and secured $31 million in seed funding led by Andreessen Horowitz. FLUX.1, available in three variants, boasts 12 billion parameters and advanced multimodal architecture. The models aim to rival closed-source alternatives like Midjourney and DALL-E. 

This launch comes amid challenges at Stability AI, reinvigorating the open-source AI community. Black Forest Labs emphasizes ethical AI use, prohibiting harmful content. The company plans to expand into text-to-video systems, aiming to transform generative media technology across various industries.

VENTUREBEAT

OpenAI settles into its VC era

OpenAI has entered a new venture capital phase, as highlighted in Fortune's Eye on AI newsletter. CEO Sam Altman is steering the company towards strategic investments, fostering growth in AI development. This shift reflects OpenAI’s commitment to expanding its influence and innovation in the AI sector. 

The company aims to balance advancing technology with ethical considerations, addressing concerns about AI's impact on society. This move comes as part of a broader trend of AI companies seeking funding and partnerships to stay competitive in a rapidly evolving market.

FORTUNE

🧠RESEARCH

Gemma 2: Improving Open Language Models at a Practical Size

Gemma 2 introduces improved lightweight open language models with 2 to 27 billion parameters. Using advancements in Transformer architecture and knowledge distillation, these models achieve superior performance relative to their size. Gemma 2 models are competitive with significantly larger models, offering efficient and powerful alternatives. All models are freely available to the community.

SAM 2: Segment Anything in Images and Videos

SAM 2 is an advanced model for visual segmentation in images and videos, utilizing a transformer architecture with streaming memory for real-time processing. It boasts the largest video segmentation dataset, achieving superior accuracy with fewer user interactions. SAM 2 is 6x faster and more accurate than its predecessor, SAM. The model, dataset, and an interactive demo are available for public use.

MindSearch: Mimicking Human Minds Elicits Deep AI Searcher 

MindSearch is a new AI framework mimicking human cognitive processes to improve web information retrieval and integration. It uses a multi-agent system to break down queries, search multiple web pages, and compile relevant information efficiently. This method significantly enhances response quality and speed, outperforming existing AI search engines like ChatGPT-Web and Perplexity.ai.

Theia: Distilling Diverse Vision Foundation Models for Robot Learning

Theia is a vision foundation model for robots that integrates multiple existing vision models to enhance robot learning. It encodes diverse visual knowledge, improving performance with less training data. Theia outperforms previous models and its own teachers in experiments, demonstrating better efficiency and effectiveness in vision-based robot policy learning.

MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts

MoMa is a new architecture for pre-training mixed-modal language models, efficiently handling images and text. It uses modality-specific expert modules, improving pre-training efficiency significantly. With a 1-trillion-token budget, MoMa achieves 3.7x overall FLOPs savings, outperforming traditional models. Combining with mixture-of-depths increases savings further but may affect causal inference performance.

🛠️TOP TOOLS

Workflows by Llamaindex - Visualize, interact with, and debug complex agent-based applications.

Simplescraper - Receive Markdown, JSON or a CSV file in seconds that you can upload directly to OpenAI, Claude and more

LangGraph Studio - Offers a new way to develop LLM applications by providing a specialized agent IDE that enables visualization, interaction, and debugging of complex agentic applications

Aiva - Your personal AI music generation assistant

Gendo - AI for Architecture & Design

📲SOCIAL MEDIA