| |
| Here's a look at today's AI briefing: - Nvidia unveils next-generation AI GPU.
- Apple in talks to use Google Gemini on iPhones.
- YouTube now requires creators to label AI.
Beth
p/beth-duckett | |
| 1 | | Nvidia today unveiled its latest and most advanced series of AI chips. CEO Jensen Huang announced the next generation of graphics processing units, based on its new Blackwell architecture, at today's GTC 2024 developer's conference. More: - The flagship GPU, Blackwell B200, succeeds the company's Hopper current generation of GPUs.
- It is the "world's most powerful" AI chip, according to Nvidia.
- It is expected to ship later this year, allowing companies to train bigger and more complex AI models.
- Blackwell is named after mathematician and statistician David Blackwell, the first Black inductee into the U.S. National Academy of Sciences.
- "Blackwell will be the most successful product launch in our history," Huang said.
By the numbers: - Blackwell is twice as powerful at AI model training and offers five times the inference capability compared to Hopper.
- Companies can use the GPU to build and run genAI in real time on trillion-parameter language models at up to 25 times lower cost and energy consumption, according to Nvidia.
- At 208 billion transistors, Blackwell is over double the count of Nvidia's previous 80 billion transistor chip, the popular Hopper H100.
- Nvidia claims training a 1.8-trillion-parameter model now requires only 2,000 Blackwell GPUs and 4 megawatts, a task that previously needed 8,000 Hopper GPUs and 15 megawatts.
Zoom out: - In addition, the GB200 "superchip" will combine two Blackwell GPUs with Nvdia's "Grace" CPU.
- An entire server, called GB200 NVLink 2, will merge 72 Blackwell GPUs and other Nvidia components.
- Amazon, Google, Microsoft, and Oracle plan to include NVL72 racks in their cloud services.
|     | |
| 2 | | Apple is in talks with Google to license its Gemini AI models for upcoming iPhone features, according to Bloomberg sources. Apple has also held talks with OpenAI and is considering using its GPT models, the sources said. More: - The "blockbuster" deal would allow Apple to use Google's Gemini AI for more advanced services like chatbots and image generators on iPhones.
- The capabilities would be a part of iOS 18, the next major version of the iPhone's operating system, due out later this year.
- If the deal falls through, Apple could choose another AI provider like OpenAI or Anthropic, according to Bloomberg.
Zoom out: - It's unclear how much Apple could pay Google for Gemini access.
- A partnership with Apple would offer a major boost to Google's AI efforts, potentially bringing Gemini to over 2 billion active Apple devices.
- An AI deal between Apple and Google would also expand their existing search partnership, where Google pays Apple billions of dollars a year to be Safari's default search engine.
New Apple models: - Apple is also developing AI-powered features for iOS 18 based on its own AI models, but they would be on-device rather than in the cloud.
- A new pre-print paper describes Apple's MM1, a family of multimodal models with up to 30B parameters.
- Large-scale pre-training equips MM1 with advanced features like improved in-context learning and multi-image reasoning.
- After supervised fine-tuning, the MM1 models were competitive on various multimodal benchmarks.
| | | |
| A message from our sponsor, DoiT. |  | | This practical guide provides over 70 self-contained recipes to help you creatively solve common AWS challenges you’ll encounter on your cloud journey. If you’re comfortable with rudimentary scripting and general cloud concepts, this cookbook provides what you need to address foundational tasks and create high-level capabilities. Go beyond theory and learn the details you need to successfully build on AWS. The recipes help you: - Redact personal identifiable information (PII) from text using Amazon Comprehend
- Automate password rotation for Amazon RDS databases
- Use VPC Reachability Analyzer to verify and troubleshoot network paths
- Lock down Amazon Simple Storage Service (S3) buckets
- Analyze AWS Identity and Access Management policies
- Autoscale a containerized service
Download the AWS Cookbook, compliments of the cloud experts at DoiT. Download Now | |
|
| 3 | | YouTube is now requiring creators to label realistic-looking AI-generated content in videos. In a blog post today, YouTube says the labels are meant to improve transparency for viewers "and build trust between creators and their audience." More: - Starting today, YouTube has a new tool in Creator Studio for creators to disclose AI content that appears realistic.
- Creators can check a box during the upload process to label videos, audio, and photos with AI-generated or synthetic content.
- The labels will begin appearing on YouTube in the "coming weeks," starting on mobile and then TV and desktop.
- Viewers will see the labels in expanded video descriptions or on the front of the video player.
- For sensitive topics like news, elections, and health, labels will be prominently displayed on the video.
When the rules apply: - The rules only apply to content mimicking real people, events, or places, including the likeness of people, altered footage, and fictional yet realistic scenes.
- They exclude captions, animations, unrealistic content, and minor AI edits like color changes or beauty effects.
| | | |
| 4 | | The U.S. Department of Homeland Security (DHS) announced three AI pilot projects today as part of its first-ever "Artificial Intelligence Roadmap." DHS agencies will use AI models, such as OpenAI's ChatGPT, and collaborate with Anthropic, Meta, Microsoft, Google, and Amazon for the pilot programs. More: - In the first test, Homeland Security Investigations will use AI to help with investigations into fentanyl detection and combat child sexual exploitation.
- Investigators will be able to use AI to quickly summarize and search through reports.
- Another DHS agency, FEMA, will use AI to help communities plan and create hazard mitigation strategies.
- In the third test, U.S. Citizenship and Immigration Services (USCIS) is using AI to improve its immigration officer training.
Zoom out: - DHS also unveiled its roadmap for AI deployment, with an emphasis on privacy, civil rights, bias prevention, and national safety, such as protecting cyber networks.
- The roadmap advocates for partnerships across the private sector, local governments, and research institutions.
| | | |
| 5 | | Over 60% of video game studios are now using AI to improve their workflows, especially for content creation and character animation, according to Unity's 2024 Gaming Report. Among game developers surveyed, the top use of AI is improving character animations, followed by coding, and then creating game levels and art. Just over 70% of the developers who use AI reported improvements in their delivery and operations. 
The survey data shows that 56% of developers using AI focus on worldbuilding, with 64% of them preferring AI to create non-playable characters for these immersive environments. 
AI is most commonly used to build AR/VR and online multiplayer games, according to the survey.
Despite AI's surge in 2023, 38% of studios polled express reluctance to adopt it. Unity's report shows that 43% of developers hesitant to use AI are interested but lack time, while 24% lack the technical skills, and the rest are unaware of AI's capabilities or purpose. 
*Unity's 2024 report draws on data from its engine and tool users and a 2023 survey by market research firm Cint with 300 participants. | | | |
| 6 | | AI Events and Conferences: - Nvidia GTC AI Conference & Expo: March 18-21, San Jose, California, and virtual.
- Imagine AI Live '24: March 27-28, Las Vegas.
- World Summit AI Americas: April 24-25, Montreal.
- Data Innovation Summit: April 24-25, Stockholm and online.
- Rise of AI Conference: May 14-15, Berlin, Germany, and virtual.
- Enterprise Generative AI Summit West Coast: May 21-22, Silicon Valley, California.
- AI Expo Tokyo: May 22-24, Tokyo.
- Data Cloud Summit 2024: June 3-6, San Francisco.
- SuperAI: June 5-6, Singapore.
- 2024 IEEE Conference on Artificial Intelligence: June 25-27, Singapore.
- The AI Conference: Sept. 18-19, San Francisco.
| | | |
| 7 | | QUICK HITS - Sign up for real-time, easy-to-understand business credit alerts and get the insights you need to help you reach your business goals.*
- Super Micro Computer joined the S&P 500 today. The company's stock has soared nearly 290% this year, driven by heavy sales of its servers equipped with Nvidia's AI chips.
- Qualcomm's new Snapdragon 8s Gen 3 processor for smartphones supports on-device AI models.
- Adobe introduced two AI-powered beta features to its 3D design software. The tools can generate 3D model textures as well as 3D scene background images from text prompts.
- Visa and Mastercard have reached a groundbreaking $5B settlement, reshaping the landscape of financial transactions as we know it.*
*This is a sponsored post. | | | |
| Upcoming Events | | MAR
27 | | Join DTC industry leaders and CEOs at the pivotal virtual roundtable event 'Shaping the Future of DTC in 2024.' | | | | | | | SEP
1 | | Join the $1M prize AI Challenge: Unleash your creativity and push AI Agents beyond today's limits - | | | | | | * This is a sponsored event | | | |
 | AI and technology writer | | Beth is a contributing editor and writer of Inside's AI and Tech newsletters. She has written for publications including USA Today, the Arizona Business Gazette, and The Arizona Republic, where she received recognition with a Pulitzer Prize nomination and a First Amendment Award for collaborative reporting on state pension cost increases. You can reach her at Beth.Duckett@yahoo.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
| |
|
| |
Comments
Post a Comment