Today:

• Microsoft, Apple Exit OpenAI Board

• Groq Nears $2.2 Billion Valuation

• AMD Matches Nvidia with Silo

• Andreessen Hoards GPUs for AI

• Stability AI Enhances Chatbot Features

• AWS Launches App Studio Tool

Microsoft and Apple ditch OpenAI board seats amid regulatory scrutiny

Microsoft and Apple have stepped down from OpenAI’s board amid regulatory scrutiny. Microsoft held a non-voting observer seat for less than eight months. OpenAI will now conduct regular stakeholder meetings instead of maintaining board seats for its partners. This move comes as regulatory concerns grow over Microsoft's $10 billion investment in OpenAI, which has made Microsoft its exclusive cloud partner. UK, EU, and FTC regulators are investigating Microsoft’s partnership with OpenAI. 

Despite the changes, OpenAI will continue to collaborate closely with key partners like Microsoft and Apple to enhance safety and security in their AI endeavors.

THE VERGE

Groq, an Nvidia Rival, Nears $2.2 Billion-Valuation BlackRock Deal; Musk’s Mystery Company Revealed

Groq, an AI chip startup, is set to raise over $300 million from BlackRock, potentially valuing the company at $2.2 billion. This funding, more than doubling its previous $1.1 billion valuation from 2021, highlights the competitive nature of the AI chip market dominated by Nvidia. 

However, BlackRock's deal requires Groq to undergo a financial audit, an uncommon demand in the fast-paced AI investment landscape but indicative of the scrutiny AI hardware startups face. Despite these challenges, Groq's specialized chips for large language models continue to attract significant investment and interest.

THE INFORMATION

AMD buys developer Silo AI in bid to match Nvidia's product range

AMD plans to acquire Finnish AI software company Silo AI for $650 million to enhance its enterprise AI offerings and compete with Nvidia. This all-cash deal, expected to close in the second half of 2024, will integrate Silo AI into AMD's Artificial Intelligence Group. Silo AI, Europe's largest private AI lab, specializes in integrating AI into products and services for clients like Philips and Rolls-Royce. 

While the acquisition won't immediately impact AMD's financials, it aims to unlock future business potential and strengthen AMD's AI capabilities against Nvidia's comprehensive AI solutions.

THE REGISTER

Andreessen Horowitz Is Building a Stash of More Than 20,000 GPUs to Win AI Deals

Venture capital firm Andreessen Horowitz is stockpiling over 20,000 GPUs, including Nvidia H100 units, to attract AI startups. The initiative, known as "oxygen," involves renting these powerful AI chips to portfolio companies, enhancing their ability to develop and scale AI technologies. 

This move aligns Andreessen Horowitz with other major players in AI, such as Elon Musk’s xAI, which uses a similar number of GPUs for its large-language model. The strategy aims to provide a competitive edge in securing deals with high-potential AI startups amidst a competitive landscape and ongoing chip shortages.

THE INFORMATION

Stability AI Releases Stable Assistant Features

Stability AI has enhanced its Stable Assistant chatbot with new features. The Search & Replace tool allows users to seamlessly replace objects in images, and Stable Audio can generate high-quality, three-minute musical tracks. Existing features include upscaling, image editing, and video creation from images. 

The platform uses Stable Diffusion 3 for advanced image generation and offers tools like Enhance for improving degraded images, Outpaint for expanding visuals, and Sketch to Image for refining hand-drawn sketches. Stable Assistant aims to provide creative professionals with powerful tools for content creation and manipulation.

STABILITY AI

AWS App Studio turns text into enterprise apps in minutes

Amazon Web Services (AWS) has unveiled App Studio, a generative AI service enabling enterprise users to create applications by describing their needs in natural language. Announced at AWS's annual summit, this tool aims to democratize app development by automating coding tasks, from testing to deployment. 

App Studio promises to produce secure, scalable applications, competing with traditional and low-code development tools. Initially available in preview in the U.S. West region, notable early users include Campus Life & Style and Deloitte. AWS plans to expand access and offer the service for free, charging only for usage time.

VENTUREBEAT

🧠RESEARCH

PaliGemma: A versatile 3B VLM for transfer

PaliGemma is a versatile Vision-Language Model (VLM) combining SigLIP-So400m vision encoder and Gemma-2B language model. It performs well across nearly 40 tasks, including standard benchmarks and specialized tasks like remote-sensing and segmentation, showcasing its effectiveness in diverse open-world applications.

LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models

LLaVA-NeXT-Interleave enhances Large Multimodal Models (LMMs) by addressing multi-image, video, 3D, and single-image scenarios. Utilizing an interleaved data format and the M4-Instruct dataset with 1.18 million samples, it excels in diverse tasks. It maintains single-image performance while introducing new capabilities like cross-setting and cross-modality task transfers.

Video-STaR: Self-Training Enables Video Instruction Tuning with Any Supervision

Video-STaR is a novel self-training method for Large Vision Language Models (LVLMs) that utilizes any labeled video dataset for instruction tuning. It cycles between generating instructions and fine-tuning, enhancing general video understanding and adapting LVLMs to new tasks. Results show significant performance improvements in video question-answering and downstream tasks.

LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages

LLaMAX enhances Large Language Models (LLMs) to support translation in over 100 languages, focusing on low-resource languages. Utilizing 35,000 GPU hours for extensive multilingual pre-training, LLaMAX improves translation performance by over 10 spBLEU points without losing generalization. It matches specialized models like M2M-100-12B on the Flores-101 benchmark. Code and models are publicly available.

MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?

MJ-Bench is a new benchmark for evaluating multimodal judges used in text-to-image models like DALLE-3 and Stable Diffusion. It assesses alignment, safety, image quality, and bias through a comprehensive dataset. Experiments show closed-source models like GPT-4o provide superior feedback. The benchmark confirms these findings through human evaluations. Data and models are publicly available.

🛠️TOP TOOLS

Veed - Free AI video editor with text to video, avatars, auto-subtitles, voice translations and more.

Virtulook AI -Virtulook is an AI-powered product photo generator that allows you to easily create high-quality photographs of your products.

Audioscribe - Messy mind? Just hit record, spell it all out, and we'll turn all that into a well structured note.

Crew - AI-powered platform that recruits, vets & pays your team globally. 

Julius - Analyze your data with computational AI.

🗞️MORE NEWS

OpenAI and Los Alamos National Laboratory announce bioscience research partnership

OpenAI and Los Alamos National Laboratory are partnering to explore safe uses of AI in bioscience. They will evaluate AI models like GPT-4o in lab settings, focusing on safety and effectiveness in real-world biological tasks. This collaboration aims to enhance scientific research while ensuring AI safety. OPENAI

France’s Bioptimus Releases AI Model for Disease Diagnosis

Bioptimus, a French startup, has launched H-optimus-0, an AI model trained on hundreds of millions of images to assist in disease research and diagnosis. This open-source model can identify cancerous cells and detect genetic abnormalities in tumors, marking a significant advancement in pathology. BLOOMBERG

Defense AI startup Helsing raises $487M Series C, plans Baltic expansion to combat Russian threat

Defense AI startup Helsing raised $487 million in a Series C round led by General Catalyst, aiming to expand into Baltic countries to counter the Russian threat. Helsing plans to invest €70 million in Baltic defense projects, enhancing AI capabilities for military applications, and supporting European defense and democracy. TECHCRUNCH

Google says Gemini AI is making its robots smarter

Google is using Gemini AI to improve its robots' ability to navigate and complete tasks. By watching video tours of environments, the robots learn to execute commands based on observed information. Initial tests show a 90% success rate in task completion, highlighting significant advancements in robot training and interaction. THE VERGE

Fine-tune Claude 3 Haiku in Amazon Bedrock

Anthropic's Claude 3 Haiku AI model can now be fine-tuned in Amazon Bedrock, allowing businesses to customize its knowledge and capabilities for specific tasks. This process improves accuracy, reduces costs, and ensures consistent, brand-aligned outputs. Companies like SK Telecom and Thomson Reuters have seen significant benefits, enhancing customer support and operational efficiency. ANTHROPIC

BNP Paribas, Mistral AI ink partnership deal

BNP Paribas has partnered with Mistral AI to implement its large language models across various business areas, including customer support, sales, and IT. Mistral AI, founded by former Google DeepMind and Meta AI scientists, aims to be a European alternative to OpenAI. The partnership aims to enhance BNP Paribas' client offerings through generative AI. ECONOMIC TIMES