GenAI Daily Report 2024-03-19

News Express

NVIDIA Launches Powerful Blackwell AI GPU

Nvidia NIM Accelerates AI Model Deployment into Production

Nvidia Unveils Project GR00T for Humanoid Robot AI

Lex Fridman's Podcast Highlights with Sam Altman

Google's Gemini: A Comprehensive Generative AI Platform

ClearML Releases Open-Source Fractional GPU Tool

Apple Considers Google's Gemini AI for iPhone Features

Elon Musk's xAI Open Sources Grok AI Model

Nvidia's GTC 2024 Event Attracts Global AI Community

Nvidia Partners with CrowdStrike and Dataloop for AI Enhancements

StabilityAI Launches Stable Video 3D for Enhanced 3D Modeling

Quilt Develops AI Assistants to Empower Solutions Teams

DHS Launches AI Pilots for Public Safety and Immigration

Microsoft to Launch Qualcomm-Powered Surface Devices in May

Industry

NVIDIA Launches Powerful Blackwell AI GPU

NVIDIA introduced Blackwell, the most powerful AI GPU to date, succeeding Hopper and significantly boosting AI compute capabilities.

Detail:

  • The DGX Grace-Blackwell GB200 achieves over 1 Exaflop compute in a single rack, marking a milestone in AI hardware.
  • Blackwell's performance enables training of GPT-4 models with 1.8T parameters in 90 days on 2000 units, using a quarter of the power compared to previous models.
  • Performance comparisons highlight Blackwell's superiority over the Hopper GPU, with improvements in various computing metrics by 2.5 to 5 times.
  • The GB200 NVL72 supercomputing product integrates 36 CPUs and 72 GPUs, offering up to 1.44 exaFLOPS of inference performance.
  • NVIDIA's new chipset arrangement is humorously likened to a robot face, illustrating the innovative design of the Blackwell GPU.

URL: https://twitter.com/DrJimFan/status/1769829758479876130 From: twitter DrJimFan

Nvidia NIM Accelerates AI Model Deployment into Production

Nvidia announced Nvidia NIM at its GTC conference, promising faster AI model deployment into production by simplifying the process with AI-ready containers.

Detail:

  • Nvidia NIM combines AI models with an optimized inferencing engine into a container, accessible as a microservice.
  • Aims to shorten development time from weeks or months to instantly, even for companies without in-house AI talent.
  • Supports models from major companies and open models, enhancing Nvidia's hardware with a software layer for AI deployment.
  • Will be integrated into Amazon SageMaker, Google Kubernetes Engine, and Azure AI, among other platforms and frameworks.
  • Features like Riva, cuOpt, and Earth-2 model for various AI applications, with plans to expand capabilities over time.
  • Nvidia collaborates with companies like Box, Cloudera, and Dropbox, aiming to transform enterprises into AI-powered entities.

URL: https://twitter.com/robhof/status/1769855366320582890 From: twitter robhof

Nvidia Unveils Project GR00T for Humanoid Robot AI

Nvidia introduces Project GR00T, a foundational AI model aimed at advancing humanoid robotics, marking a significant step in AI-driven robotic development.

Detail:

  • Project GR00T announced at GTC 2024, focusing on creating a general-purpose foundation model for humanoid robots.
  • GR00T aims to enable robots to understand and execute multimodal instructions, learning from human demonstrations.
  • The initiative is part of Nvidia's broader effort to lead in AI and robotics, leveraging its hardware and software advancements.
  • Collaborations with leading humanoid robot companies are underway to ensure GR00T's wide applicability and integration.
  • Nvidia's new hardware, including Jetson Thor, supports GR00T, showcasing significant advancements in AI processing and robotics simulation.

URL: https://twitter.com/DrJimFan/status/1769892846482702658 From: twitter DrJimFan

YouTube Mandates AI Content Disclosure, DHS and Roblox Embrace AI

Major entities like YouTube, DHS, and Roblox are implementing significant AI integrations, reflecting the growing influence of AI technology.

Detail:

  • YouTube introduces a policy requiring creators to disclose AI-generated realistic content to maintain transparency.
  • The Department of Homeland Security announces a $5 million AI plan for border security, marking its extensive integration of AI for various operations.
  • Roblox unveils AI-powered tools for 3D content creation, aiding creators in generating textures and automating model rigging.
  • Nvidia reveals Blackwell GPUs at the GTC conference, showcasing a significant leap in AI accelerator technology.
  • OpenAI's Sora & Open-Sora 1.0 introduces an open-source text-to-video model, aiming to revolutionize video creation.

URL: https://twitter.com/chiefaioffice/status/1769856096540258569 From: twitter chiefaioffice

Nvidia GPU Architecture Powers Trillion-Parameter AI Models

Nvidia's GPU architecture is set to power a new generation of generative AI models with trillion parameters, promises significant advancements.

Detail:

  • Nvidia's GPU architecture will be used in developing generative AI models.
  • These AI models will have trillion parameters, indicating a leap in complexity and capability.
  • The advancement is expected to drive significant progress in AI applications.

URL: https://twitter.com/robhof/status/1769828963428278308 From: twitter robhof

Lex Fridman's Podcast Highlights with Sam Altman

Lex Fridman and Sam Altman discuss AI developments, lawsuits, and future technologies in a revealing podcast.

Detail:

  • Lex Fridman interviews Sam Altman, covering topics like OpenAI's board legends, Elon Musk's lawsuit, Ilya Sutskever's whereabouts, and the speculative existence of 'Q'.
  • They delve into AI advancements and ethics, with specific discussions on Sora's understanding of world models and the secrecy surrounding GPT-5's release.
  • Altman denies rumors of a massive $7 trillion fundraising and the existence of 'Q', adding mystique to their ventures.
  • Fridman and Altman also touch on the potential risks and ethical considerations of powerful AI systems, emphasizing the importance of safety in AI development.

URL: https://twitter.com/FinanceYF5/status/1769856401831305390 From: twitter FinanceYF5

Google's Gemini: A Comprehensive Generative AI Platform

Gemini, Google's newest generative AI platform, offers multimodal capabilities but faces mixed early reviews.

Detail:

  • Gemini is developed by Google's AI labs, DeepMind and Google Research, featuring multimodal models trained on diverse data including audio, images, videos, and text.
  • Gemini includes three models: Gemini Ultra, Gemini Pro, and Gemini Nano, each designed for different levels of complexity and applications.
  • Gemini Ultra supports tasks like physics homework help and scientific research, available through Google One AI Premium Plan at $20/month.
  • Gemini Pro, improved in its 1.5 version, excels in reasoning and understanding, available via Vertex AI for developers with specific customization options.
  • Gemini Nano, optimized for efficiency, powers features on Google's Pixel 8 Pro, offering functionalities like summarizing audio recordings and smart reply in messaging.
  • Despite Google's claims of Gemini's superiority in benchmarks, early impressions reveal issues with accuracy, translations, and code suggestions.
  • Gemini's accessibility varies, with some models free in preview while others require subscription, and its full potential and pricing details are yet to be fully unveiled.

URL: https://techcrunch.com/2024/03/18/what-is-google-gemini-ai/ From: techcrunch Kyle Wiggers

ClearML Releases Open-Source Fractional GPU Tool

ClearML has introduced an open-source tool for fractional GPU usage and new monitoring features, enhancing AI development efficiency.

Detail:

  • ClearML, an AI development platform, has launched a new open-source tool allowing fractional GPU usage.
  • The update includes advanced monitoring features designed to improve the efficiency of AI development and deployment.
  • This development is aimed at providing more accessible and efficient resources for developers and companies involved in AI projects.

URL: https://twitter.com/robhof/status/1769855489511448856 From: twitter robhof

Apple Considers Google's Gemini AI for iPhone Features

Apple is in discussions with Google to use the Gemini AI model to enhance iPhone functionalities, amidst competition in the AI space.

Detail:

  • Apple is reportedly negotiating with Google to incorporate the Gemini AI model into iPhone features, aiming to introduce AI-powered enhancements with upcoming iOS updates.
  • The talks with Google and potentially OpenAI for GPT models indicate Apple's efforts to catch up in AI, as competitors like OpenAI and Microsoft advance.
  • Apple's own AI developments might be included in iOS 18, but the company explores third-party AI for generative tasks like image creation and writing assistance.
  • Despite issues with Gemini, including paused features due to inaccuracies and blocked election-related queries, Google's experience in smartphone AI with Samsung and its Pixel phones positions it as a strong partner for Apple.

URL: https://twitter.com/robhof/status/1769755661074870602 From: twitter robhof

Elon Musk's xAI Open Sources Grok AI Model

Elon Musk's xAI has made its Grok AI model open source, marking a significant move in the AI development community.

Detail:

  • xAI released the base code of Grok AI model, excluding training code, under Apache License 2.0.
  • Grok-1, a 314 billion parameter model, was not tuned for specific applications and lacks details on its custom training stack.
  • The open source release does not include connections to X social network, contrasting its previous chatbot form for Premium+ users.
  • Other companies, like Perplexity, plan to fine-tune Grok for specific applications, such as conversational search.
  • Musk's decision to open source Grok follows a legal battle with OpenAI, reflecting a broader rivalry and his commitment to open source development.

URL: https://techcrunch.com/2024/03/18/why-elon-musks-ai-company-open-sourcing-grok-matters-and-why-it-doesnt/ From: techcrunch Devin Coldewey,@techcrunch

Nvidia's GTC 2024 Event Attracts Global AI Community

Nvidia's GTC event this week is a major gathering for the AI industry, featuring a keynote from CEO Jensen Huang.

Detail:

  • Nvidia, primarily known for gaming hardware, now leads in AI hardware market.
  • CEO Jensen Huang to deliver a keynote at the GTC 2024 event.
  • Many AI companies and startups using Nvidia technology expected to participate.
  • TechCrunch to provide ongoing coverage, highlighting the event's significance in the AI sector.

URL: https://techcrunch.com/2024/03/18/techcrunch-minute-why-the-ai-world-is-gathering-at-nvidias-gtc-2024-event-this-week/ From: techcrunch Alex Wilhelm,Yashad Kulkarni

Nvidia Partners with CrowdStrike and Dataloop for AI Enhancements

Nvidia collaborates with CrowdStrike and Dataloop to boost cybersecurity and AI development for businesses.

Detail:

  • Nvidia and CrowdStrike form a strategic partnership to improve cybersecurity through AI.
  • Nvidia collaborates with Dataloop to enhance AI application development for business environments.

URL: https://twitter.com/robhof/status/1769877289695920542 From: twitter robhof

Nvidia Launches Cloud Service for Quantum Simulations

Nvidia's latest cloud service aims to boost quantum computing simulations and enhance post-quantum security.

Detail:

  • Nvidia introduces a new cloud service designed to accelerate quantum computing simulations.
  • The service also focuses on strengthening post-quantum security.
  • Announcement covered by SiliconANGLE.

URL: https://twitter.com/robhof/status/1769856002965545117 From: twitter robhof

StabilityAI Launches Stable Video 3D for Enhanced 3D Modeling

StabilityAI introduces Stable Video 3D, advancing 3D modeling technology with improved quality and multi-view capabilities.

Detail:

  • Stable Video 3D is a generative model based on Stable Video Diffusion, aiming to progress in the 3D technology field.
  • It outperforms open-source alternatives like Zero123-XL and the previous stable version of Zero123 in model performance.
  • The model supports single image input for track video generation without camera adjustments (SV3D_u) and extends to accommodate single images and track views for creating 3D videos along specified camera paths (SV3D_p).
  • Stable Video 3D is now available for both commercial and non-commercial use through Stability AI membership.

URL: https://twitter.com/FinanceYF5/status/1769859486167302557 From: twitter FinanceYF5

Quilt Develops AI Assistants to Empower Solutions Teams

Quilt, co-founded by Dan Chen and Michael Graczyk, aims to enhance solutions teams' efficiency with AI-powered assistants tailored for tasks like filling out requests for proposals.

Detail:

  • Quilt's AI assistants are designed to support solutions engineers by automating routine tasks such as completing security questionnaires and prepping for demos.
  • The platform leverages generative AI to understand context and incorporate engineers' technical knowledge, aiming to save time and help win more deals.
  • Despite concerns about generative AI's inaccuracies, Quilt asserts its models are less prone to 'hallucinations' by separating known facts from enterprise data.
  • Addressing privacy and security risks, Quilt promises not to share data across organizations and offers users control over their data.
  • Backed by a $2.5 million seed round led by Sequoia, Quilt plans to expand its team and further develop its AI solutions for sales organizations.

URL: https://techcrunch.com/2024/03/18/quilt-is-building-ai-assistants-for-solutions-teams/ From: techcrunch Kyle Wiggers

DHS Launches AI Pilots for Public Safety and Immigration

The Department of Homeland Security is implementing AI pilots to improve public safety and streamline immigration procedures.

Detail:

  • DHS is adopting artificial intelligence to enhance public safety measures.
  • AI technology is also being utilized to make immigration processes more efficient.

URL: https://twitter.com/robhof/status/1769876826598653953 From: twitter robhof

Microsoft to Launch Qualcomm-Powered Surface Devices in May

Microsoft is set to introduce new Surface devices powered by Qualcomm in May, according to reports.

Detail:

  • Microsoft plans to unveil new Surface devices.
  • These devices will be powered by Qualcomm processors.
  • The launch is scheduled for May.

URL: https://twitter.com/robhof/status/1769863654969893021 From: twitter robhof

Investment

Splunk

Description: Splunk is a powerhouse in data analysis, security, and observability tools. It provides AI-driven solutions to help organizations connect and protect every aspect of their operations.
Amount: $28 Billion, Round: Acquisition
URL: https://www.securityweek.com/cisco-completes-28-billion-acquisition-of-splunk/

Run:ai

Description: Run:ai is an AI infrastructure startup founded in 2018 by Omri Geller and Ronen Dar. They provide a Kubernetes-based container platform for AI clouds, optimizing GPU resources for faster deep learning model training.
Amount: $1B, Round: Series C
URL: https://siliconangle.com/2024/03/18/report-nvidia-pay-1b-acquire-ai-infrastructure-startup-runai/

Zephyr AI

Description: Zephyr AI is a healthcare technology company focused on democratizing precision medicine through explainable AI algorithms. They leverage AI to improve patient stratification, response predictions, and real-world data federation in oncology and cardiometabolic disease.
Amount: $111M, Round: Series A
URL: https://www.businesswire.com/news/home/20240312475536/en/Zephyr-AI-Raises-111-Million-in-Series-A-Financing

Bear Robotics

Description: Bear Robotics pioneers in service robotics and artificial intelligence solutions for smart warehousing and supply chain automation, with a focus on autonomous navigation systems and adaptive learning algorithms, aiming to revolutionize efficiency and productivity in modern supply chains and manufacturing processes.
Amount: $60M, Round: Series C
URL: https://www.bearrobotics.ai/blog/bear-robotics-secures-60m-series-c-funding-led-by-lg-electronics

Figure Markets

Description: Figure Markets is revolutionizing private capital markets through blockchain technology, offering a decentralized financial marketplace for global trading and investment. Led by Mike Cagney and June Ou, the platform integrates MPC technology to enhance security and control over assets, aiming to reshape digital asset trading.
Amount: $60M, Round: Series A
URL: https://www.businesswire.com/news/home/20240317373347/en/Figure-Technologies-Announces-Figure-Markets-Home-to-a-New-Decentralized-Custody-Crypto-Exchange-and-Blockchain-Native-Security-Marketplace

Ocient

Description: Ocient is a data analytics software solutions company that enables always-on, compute-intensive analysis of complex, large-scale data with outstanding performance. It delivers up to 80% price savings and offers data transformation, loading, complex query processing, AI, OcientML™ and OcientGeo™ in a single solution for deeper insights and data-driven decision making.
Amount: $49.4M, Round: Series B extension
URL: https://ocient.com/newsroom/press-releases/ocient-secures-49-4-million-to-power-the-growth-of-its-energy-efficient-data-analytics-solutions/

Empathy

Description: Empathy is a technology company revolutionizing grief support and the loss of loved ones. They provide comprehensive assistance through life insurance benefits and employer bereavement leave, impacting 5 million employees and 35 million policyholders. Empathy's $90 million funding solidifies its position in the compassionate economy.
Amount: $47M, Round: Series B
URL: https://www.globenewswire.com/news-release/2024/03/12/2844600/0/en/Empathy-Announces-47M-Series-B-Solidifying-Its-Position-as-a-Leader-in-the-Emerging-Compassionate-Economy.html

CodaMetrix

Description: CodaMetrix provides an AI-powered platform for multi-specialty medical coding, partnering with leading healthcare systems to achieve autonomous coding, reduce costs, and improve provider satisfaction.
Amount: $40M, Round: Series B
URL: https://www.codametrix.com/codametrix-announces-40m-series-b-financing/

Unstructured

Description: Unstructured Technologies Inc. pioneers converting unstructured data into formats readable by large language models (LLMs), enabling organizations to access and utilize over 80% of stored data efficiently.
Amount: $40M, Round: Series B
URL: https://siliconangle.com/2024/03/14/ai-focused-big-data-startup-unstructured-raises-40m-make-data-llm-ready/

HiLabs

Description: HiLabs provides AI-powered solutions to manage dirty data in healthcare, offering the MCheck platform to ingest, cleanse, and enrich critical healthcare information. The company's technology reduces operational costs and improves patient outcomes by enabling data-driven decision-making.
Amount: $39M, Round: Series B
URL: https://www.prnewswire.com/news-releases/hilabs-announces-closing-of-39-million-series-b-financing-to-advance-its-ai-powered-data-management-solutions-for-healthcare-organizations-302088742.html

Tavus Dev Platform

Description: Tavus Dev Platform raised funds for its GenAI Replica API, powered by the Phoenix model, creating lifelike videos from text for sales and education.
Amount: $18M, Round: Series A
URL: https://www.forbes.com/sites/charliefink/2024/03/14/adaptive-ais-stealthy-20-million-tavus-dev-platform-raises-18m/

Products

Top Products

EVM Sandbox Your production-ready Web3 staging environment
Pioneering enterprise-grade Web3 development staging, enhancing CI/CD pipelines for Web3 with privacy and performance.
URL: https://www.producthunt.com/r/PJS3KKNTJIJUUI?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+info.ai+%28ID%3A+116042%29

Bigship Start shipping your heavy load @Rs.*7.5/per kg
Innovative AI-based courier aggregation with cost-effectiveness and enhanced shipping features for Pan India reach.
URL: https://www.producthunt.com/r/XS7TUUNQCN2LMP?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+info.ai+%28ID%3A+116042%29

Double A Framework: AI-Driven Innovation Navigate AI with confidence. Agility meets human-centricity
Facilitates AI adoption with a human-centric approach, addressing technical uncertainties and driving business value.
URL: https://www.producthunt.com/r/MSJL4RZDTV3LAW?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+info.ai+%28ID%3A+116042%29

Spot Instances Availability Map Spot Instance availability heatmap across cloud providers
Provides a global heatmap for Spot instance availability across cloud providers, optimizing cost and reliability.
URL: https://www.producthunt.com/r/SEFMU6BQSWSYWJ?utm_campaign=producthunt-api&utm_medium=api-v2&utm_source=Application%3A+info.ai+%28ID%3A+116042%29

Research

Uni-SMART: Universal Science Multimodal Analysis and Research Transformer

Authors: Hengxing Cai, Xiaochen Cai, Shuwen Yang, Jiankun Wang, Lin Yao, Zhifeng Gao, Junhan Chang, Sihang Li, Mingjun Xu, Changxin Wang, Hongshuai Wang, Yongge Li, Mujie Lin, Yaqi Li, Yuqi Yin, Linfeng Zhang, Guolin Ke
Institution: DP Technology, AI for Science Institute, Beijing
Uni-SMART is a new model designed to understand and analyze scientific literature, including texts and multimodal elements like charts. It outperforms traditional text-focused models in understanding complex scientific information, showing promise for applications such as patent infringement detection. This innovation could significantly enhance how researchers engage with scientific literature.
Link: https://arxiv.org/abs/2403.10301

VideoAgent: Long-form Video Understanding with Large Language Model as Agent

Authors: Xiaohan Wang, Yuhui Zhang, Orr Zohar, Serena Yeung-Levy
Institution: Stanford University
This study introduces 'VideoAgent', a new system that uses a large language model to understand long videos by acting like a human would: asking questions and looking for answers. It works by picking out key information from videos with the help of other models that can understand both pictures and words. Tested on two difficult benchmarks, VideoAgent outperformed existing methods, showing it's not only more effective but also more efficient in understanding long videos.
Link: https://arxiv.org/abs/2403.10517

Recurrent Drafter for Fast Speculative Decoding in Large Language Models

Authors: Aonan Zhang, Chong Wang, Yi Wang, Xuanyu Zhang, Yunfei Cheng
Institution: Apple
This study introduces a new, efficient method for speculative decoding in large language models, blending the best of two previous techniques into a simpler, single-model strategy. It uses a lightweight, recurrent draft head for quick filtering, proving to be effective through tests on various models and analyzing its practical trade-offs.
Link: https://arxiv.org/abs/2403.09919

Alignment Studio: Aligning Large Language Models to Particular Contextual Regulations

Authors: Swapnaja Achintalwar, Ioana Baldini, Djallel Bouneffouf, Joan Byamugisha, Maria Chang, Pierre Dognin, Eitan Farchi, Ndivhuwo Makondo, Aleksandra Mojsilovi´c, Manish Nagireddy, Karthikeyan Natesan Ramamurthy, Inkit Padhi, Orna Raz, Jesus Rios, Prasanna Sattigeri, Moninder Singh, Siphiwe Thwala, Rosario A. Uceda-Sosa, Kush R. Varshney
Institution: IBM Research
This paper introduces Alignment Studio, a new method allowing developers to customize large language models to fit specific values, norms, and legal requirements. It details an architecture comprising Framers, Instructors, and Auditors to manage model behavior, demonstrated through a case study of adapting a chatbot to a company's guidelines.
Link: https://arxiv.org/abs/2403.09704

RAFT: Adapting Language Model to Domain Specific RAG

Authors: Tianjun Zhang, Shishir G. Patil, Naman Jain, Sheng Shen, Matei Zaharia, Ion Stoica, Joseph E. Gonzalez
Institution: UC Berkeley
This paper introduces RAFT, a method for enhancing Large Language Models (LLMs) to better answer domain-specific questions by teaching them to focus on relevant information and ignore distractions. RAFT improves performance by using a retrieval-augmented approach and has been proven effective across various datasets. It offers a new way to fine-tune pre-trained models for specific tasks.
Link: https://arxiv.org/abs/2403.10131