OpenAI-Musk Lawsuit

PLUS: Snowflake Integrates Mistral AI Models, Haiper Challenges AI Video Giants and more.

Today:

BREAKING: OpenAI Reveals the TRUTH About Elon Musk's Lawsuit | Sam Altman, Ilya Sutskever and more.

Elon Musk got into a tiff with OpenAI, wanting access to classified docs. Sam Altman, CEO of OpenAI, had been silent on the lawsuit Musk filed but eventually responded with a reminder of past support for Tesla and caution against betting against Musk. OpenAI published a blog detailing their relationship with Musk, emphasizing their commitment to the AGI mission for humanity's benefit. 

OpenAI chose to remain independent, focusing on building beneficial AI tools and contributing to societal improvements. The dispute reflects differing visions for AI's future and OpenAI's determination to progress without Musk, showcasing a clash between two titans of technology over control and vision for artificial intelligence's path forward.

Microsoft’s new Orca-Math AI outperforms models 10x larger

Microsoft unveiled Orca-Math, a small yet powerful AI model designed to excel in solving math word problems. Despite its modest size of 7 billion parameters, Orca-Math outperforms larger models by a factor of ten, matching the performance of Google's Gemini Ultra and OpenAI's GPT-4. The team behind Orca-Math utilized a novel approach, generating a dataset of 200,000 math problems using specialized AI agents. They also leveraged the Kahneman-Tversky Optimization method to enhance the model's accuracy. 

This groundbreaking achievement demonstrates the potential of smaller AI models to deliver impressive results, easing concerns about model scalability. Additionally, Microsoft has made the dataset openly available, encouraging further innovation in AI.

Cloudflare announces new defensive AI products for protecting LLMs and companies from attack

Cloudflare unveiled new defensive AI tools, including Firewall for AI, aimed at safeguarding large language models (LLMs) from cyber threats. With the rise of AI, phishing scams have become more sophisticated, but Cloudflare's Defensive AI can detect email threats quicker. The company also offers personalized defense strategies tailored to individual network behaviors. 

By deploying AI against AI, Cloudflare aims to enhance security and protect against evolving cyber threats. CEO Matthew Prince emphasized the importance of securing AI-powered applications, ensuring a safer internet for all. This move addresses growing concerns regarding prompt injection attacks and other emerging attack vectors, offering free protection to Cloudflare's customers.

Snowflake Partners with Mistral AI to Bring Industry-Leading Language Models to Enterprises Through Snowflake Cortex

Snowflake and Mistral AI join forces to bring powerful language models to businesses via Snowflake Cortex, enhancing data analytics and AI app development. Mistral AI’s latest model, Mistral Large, excels in language tasks and offers multilingual capabilities. Snowflake customers gain access to Mistral's top-performing models, including Mixtral 8x7B and Mistral 7B, optimizing analytics and AI app creation. 

With Snowflake Cortex LLM Functions in public preview, users can securely leverage generative AI with enterprise data. This partnership marks a significant step in empowering enterprises with advanced AI capabilities while ensuring data security and privacy.

Competition in AI video generation heats up as DeepMind alums unveil Haiper

In the competitive world of AI video creation, Haiper emerges as a contender, founded by ex-DeepMind experts Yishu Miao and Ziyu Wang. With $13.8 million in funding, Haiper offers a user-friendly platform for generating short videos using AI. While currently focused on consumer tools, Haiper aims to develop a robust video-generation model for wider use. 

Facing competition from giants like OpenAI and Google, Haiper's success hinges on overcoming challenges like the uncanny valley effect. Despite the uphill battle, investors see potential, emphasizing the need for Haiper to innovate to stand out in the burgeoning AI video market.

🧠RESEARCH

RT-H is a robot learning approach that uses language to simplify complex tasks into understandable steps. It focuses on teaching robots the "language of actions" for precise movements like "move arm forward," bridging the gap between high-level tasks and physical actions. This method enables robots to learn from and adapt to human instructions mid-task, making them more versatile and effective in multitasking environments. RT-H stands out by learning from language interventions, outperforming traditional methods reliant on direct control.

MovieLLM is a framework that creates AI-generated movies for enhancing long video understanding. This approach uses GPT-4 and text-to-image models to produce detailed scripts and visuals, offering a scalable solution to the challenge of acquiring high-quality, diverse video data. MovieLLM's synthetic data boosts multimodal models' performance, enabling them to grasp complex video narratives better by overcoming the limitations of current datasets related to scarcity and bias.

ResAdapter is a new tool that lets diffusion models create images with any resolution or aspect ratio, fixing a big problem where these models couldn't handle images outside their usual sizes. ResAdapter learns how to do this by understanding the basics of resolution itself and works with personalized models without messing up their unique styles. It's a game-changer because it works with any diffusion model, supports a wide range of resolutions, and is efficient, requiring only a small addition to existing setups.

AtomoVideo is a cutting-edge framework for turning images into high-quality videos. It uses multi-granularity image injection to ensure the videos closely match the original images while ensuring smooth motion and temporal consistency. This approach not only supports long sequence prediction but also allows integration with existing models for personalized and controllable outputs. AtomoVideo stands out by delivering superior performance compared to existing methods, proving its effectiveness through both quantitative and qualitative assessments.

"DenseMamba," introduces a novel approach to improve Large Language Models (LLMs) by enhancing the flow of information between layers in State Space Models (SSMs). By integrating hidden states from shallow layers into deeper ones, DenseMamba retains crucial fine-grained information, boosting performance. This method maintains the efficiency of training and inference, showing significant improvements in accuracy on benchmarks, exemplified by DenseRetNet's up to 5% accuracy improvement over RetNet.

🛠️TOP TOOLS

Reor - desktop app designed to transform note-taking with AI, organizing your thoughts by linking related ideas, answering questions, and providing semantic searches, all locally on your device.

Globe - enabling users to explore and interact with various geospatial data layers.

Loud Fame - offers a service to stylize videos with AI filters, including an anime style. 

DrLambda -  robo-assistant for pro slides and posts

Folk - All-in-one CRM

Smartwriter - Generate 1000s of Personalised Emails that get you 8x more replies within minutes using AI

Retell AI - Conversational Voice API for Your LLM

Longshot - AI Assistant designed specifically for content teams, offering tools for research, generation, optimization of content, and ensuring factual accuracy.

📲SOCIAL MEDIA

🗞️MORE NEWS

AMD introduces Spartan UltraScale+ programmable logic chips for edge applications

AMD unveiled its Spartan UltraScale+ programmable chips, aimed at edge applications, boasting high I/O counts, energy efficiency, and top-notch security features. These FPGAs cater to various sectors like healthcare, robotics, and video apps. Sporting advanced technology, they offer the industry's highest I/O to logic cell ratio and up to 30% lower power consumption compared to predecessors. Security-wise, they incorporate Post-Quantum Cryptography and a unique fingerprint function for enhanced protection. Supported by AMD's Vivado Design Suite, these chips are slated for sampling in 2025, reflecting AMD's commitment to innovation and addressing evolving market needs. VENTUREBEAT

CrowdStrike and Dell unleash an AI-powered, unified security vision

Dell and CrowdStrike team up to combat cyber threats using AI, merging CrowdStrike’s Falcon XDR platform with Dell's MDR service. With a surge in AI-based attacks, this collaboration offers businesses a potent defense. Partnerships like this are crucial in an ever-evolving security landscape. The future of cybersecurity lies in AI-human synergy for effective threat detection and response. VENTUREBEAT

Kayak’s new AI features will let users double-check flights with a screenshot

Kayak introduces AI features for easier travel planning, including PriceCheck to compare flight prices via screenshot and Ask Kayak, a chatbot for travel advice. Utilizing computer vision, PriceCheck scans for cheaper options, while Ask Kayak assists users in finding destinations, hotels, and activities. Expedia also explores AI for enhanced travel services. THE VERGE

AMD Hits US Roadblock in Selling AI Chip Tailored for China

The US Commerce Department bars Advanced Micro Devices (AMD) from exporting its powerful AI chip to China without a license, citing national security concerns. AMD hoped for approval due to its chip's lower performance in China, but officials insist on a license. This move aligns with US restrictions on advanced technology exports. BLOOMBERG

AI Search Startup Perplexity Set to Double Valuation to $1 Billion

Perplexity, an AI startup, is poised to double its valuation to $1 billion in a new funding round, challenging Google's dominance in web search. The surge in funding reflects investor optimism about AI's potential to disrupt the search market and revitalize startups. THE WALL STREET JOURNAL

Move over Gemini, open-source AI has video tricks of its own

A new open-source AI model, LWM, rivals Google's Gemini in video processing tasks, such as identifying frames matching text descriptions. Developed by UC Berkeley researchers, LWM outperforms Gemini 1.0 and GPT-4 Turbo while running on less computing power. Its training approach, "Ring Attention," allows for efficient parallelized training on progressively longer data sequences. ZDNET

Top AI researchers say OpenAI, Meta and more hinder independent evaluations

Top AI researchers criticize companies like OpenAI and Meta for hindering independent evaluations of their AI systems. They argue that strict company policies discourage safety-testing and transparency, leading to a chilling effect on research. The researchers call for legal and technical safeguards to enable independent scrutiny of AI tools. THE WASHINGTON POST

Alibaba joins Silicon Valley peers like Microsoft in placing big bets on generative AI, leading $600 million financing round for Chinese AI startup MiniMax

Alibaba, a major player in Asian tech finance, leads a $600 million funding round for MiniMax, a Chinese AI startup, its second major deal in 2024. With MiniMax valued at over $2.5 billion, Alibaba aims to revitalize its growth amidst regulatory scrutiny, joining peers like Microsoft in investing in generative AI. FORTUNE

New AI model accurately identifies tumors and diseases in medical images 

Researchers at the Beckman Institute for Advanced Science and Technology have developed an AI model capable of accurately identifying tumors and diseases in medical images. This model, unlike traditional black-box AI systems, provides transparent explanations for its diagnoses through visual maps, enhancing doctor-patient trust and facilitating better understanding of medical decisions. MEDICAL LIFE SCIENCES NEWS

Using generative AI to improve software testing

DataCebo, a spinout from MIT, utilizes generative artificial intelligence (AI) to create synthetic data that mirrors real-world scenarios, aiding organizations in tasks such as software testing and machine learning model training. This synthetic data enables companies to maintain privacy while preserving statistical relationships between data points and simulate various scenarios for testing purposes. The Synthetic Data Vault (SDV), developed by DataCebo, has been widely adopted, with over 1 million downloads and more than 10,000 data scientists using the open-source library. By harnessing generative AI, DataCebo aims to revolutionize software testing and enhance transparency and responsibility in data-driven decision-making processes. MIT

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.