AI Weekly Rundown (April 20 to April 26)
Major AI announcements from Apple, Microsoft, Adobe, Snowflake, Meta, and more.
Hello Engineering Leaders and AI Enthusiasts!
Another eventful week in the AI realm. Lots of big news from huge enterprises.
In today’s edition:
🍎 iOS 18 to have AI features with on-device processing
🧠 Many-shot ICL is a breakthrough in improving LLM performance
⚡ Groq shatters AI inference speed record with 800 tokens/second on LLaMA 3
📱 Microsoft launches its smallest AI model that can fit on your phone
🎨 Adobe survey says 50% of Americans use generative AI every day
👨💻 Microsoft hired former Meta VP of infrastructure
🖼️ Firefly 3: Adobe’s best AI image generation model to date
👓 Meta finally rolls out multimodal AI capabilities for its smart glasses
🧬 Profulent’s OpenCRISPR-1 can edit the human genome
🤖 NVIDIA acquires Run:ai; integrates it with DGX Cloud AI Platform
❄️ Snowflake enters the generative AI arena with Arctic LLM
⏳ Monetizing generative AI to take time, says Zuckerberg
🏭 Sanctuary AI launches Phoenix 7 robot for industrial automation
🛑 AI integration hits roadblocks for CIOs
💊 Moderna and OpenAI partner to accelerate drug development
Let’s go!
iOS 18 to have AI features with complete on-device processing
Apple is set to make significant strides in artificial intelligence with the upcoming release of iOS 18. According to AppleInsider’s recent report, the tech giant is focusing on privacy-centric AI features that will function entirely on-device, eliminating the need for cloud-based processing or an internet connection. This approach addresses concerns surrounding AI tools that rely on server-side processing, which have been known to generate inaccurate information and compromise user privacy.
The company is reportedly developing an in-house LLM called "Ajax," which will power AI features in iOS 18. Users can expect improvements to Messages, Safari, Spotlight Search, and Siri, with basic text analysis and response generation available offline. We'll learn more about Apple's AI plans at the Worldwide Developers Conference (WWDC) starting June 10.
Many-shot-in-context learning is a breakthrough in improving LLM performance
A recent research paper has introduced a groundbreaking technique that enables LLMs to significantly improve performance by learning from hundreds or thousands of examples provided in context. This approach, called many-shot in-context learning (ICL), has shown superior results compared to the traditional few-shot learning method across a wide range of generative and discriminative tasks.
To address the limitation of relying on human-generated examples for many-shot ICL, the researchers explored two novel settings: Reinforced ICL, which uses model-generated chain-of-thought rationales instead of human examples, and Unsupervised ICL, which removes rationales from the prompt altogether and presents the model with only domain-specific questions.
Both approaches have proven highly effective in the many-shot regime, particularly for complex reasoning tasks. Furthermore, the study reveals that many-shot learning can override pretraining biases and learn high-dimensional functions with numerical inputs, unlike few-shot learning, showcasing its potential to revolutionize AI applications.
Groq shatters AI inference speed record with 800 tokens/second on LLaMA 3
AI chip startup Groq recently confirmed that its novel processor architecture is serving Meta's newly released LLaMA 3 large language model at over 800 tokens per second. This translates to generating about 500 words of text per second - nearly an order of magnitude faster than the typical speeds of large models on mainstream GPUs. Early testing by users seems to validate the claim.
Groq's Tensor Streaming Processor is designed from the ground up to accelerate AI inference workloads, eschewing the caches and complex control logic of general-purpose CPUs and GPUs. The company asserts this "clean sheet" approach dramatically reduces the latency, power consumption, and cost of running massive neural networks.
Microsoft launches its smallest AI model that can fit on your phone
Microsoft launched Phi-3-Mini, a 3.8 billion parameter language model, as the first of three small models in the Phi-3 series. It is trained on a smaller dataset than larger LLMs like GPT-4 and outperforms models like Meta's Llama 2 7B and GPT-3.5 on benchmarks like MMLU and MT-bench. The Phi-3 series also includes Phi-3-Small (7B parameters) and Phi-3-Medium (14B parameters), which are more capable than Phi-3-Mini.
What sets Phi-3-Mini apart is its ability to run locally on mobile devices like the iPhone 14, thanks to its optimized size and innovative quantization techniques. Microsoft's team took inspiration from how children learn, using a "curriculum" approach to train Phi-3 on synthetic "bedtime stories" and simplified texts. While robust for its size, Phi-3-Mini is limited in storing extensive factual knowledge and is primarily focused on English.
Adobe survey says 50%Americans use generative AI everyday
Adobe surveyed 3,000 consumers on February 15-19, 2024, about their usage of generative AI and found over half of Americans have already used generative AI. The majority believe it helps them be more creative. Adobe's Firefly has generated 6.5 billion images since its inception last March. Americans use generative AI for research, brainstorming, creating content, searching, summarization, coding, and learning new skills.
Moreover, 41% of Americans expect brands to use AI for personalized shopping, price comparisons, and customer support. Adobe's data also reveals that online traffic to retail and travel sites has surged, with faster customer service and more creative experiences due to generative AI tools.
Microsoft hired former Meta VP of infrastructure
With the recent addition of Google DeepMind co-founder Mustafa Suleyman to lead Microsoft's consumer AI division, Microsoft has once again poached a former Meta VP of infrastructure. This strategic hire comes amidst rumors of Microsoft and OpenAI's plans to construct a $100 billion supercomputer, "Stargate," to power their AI models.
Jason Taylor oversaw infrastructure for AI, data, and privacy in Meta. He will join Microsoft as the corporate vice president and deputy CTO, tasked with building systems to advance the company's AI ambitions.
Firefly 3: Adobe’s best AI image generation model to date
Adobe has announced a major update to its AI image generation technology called Firefly Image 3. The model showcases a significant improvement in creating more realistic and high-quality images over previous versions. It has enhanced capabilities to understand longer text prompts, generate better lighting, and depict subjects like crowds and human expressions. The Firefly Image 3 model is now available through Adobe's Firefly web app as well as integrated into Adobe Photoshop and InDesign apps.
It powers new AI-assisted features in these apps, such as generating custom backgrounds, creating image variations, and enhancing detail. Adobe has also introduced advanced creative controls like Structure Reference to match a reference image's composition and Style Reference to transfer artistic styles between images. Adobe also attaches "Content Credentials" to all Firefly-generated assets to promote responsible AI development.
Meta finally rolls out multimodal AI capabilities for its smart glasses; adds new features
Meta has announced exciting updates to their Ray-Ban Meta smart glasses collection. They are introducing new styles to cater to a wider range of face shapes. The new styles include the vintage-inspired Skyler frames, designed for smaller faces, and the Headliner frames with a low bridge option. It also introduces video calling capabilities via WhatsApp and Messenger, allowing users to share their views during a video call.
Meta is integrating its AI technology, Meta AI Vision, into Ray-Ban smart glasses. Users can interact with the glasses using voice commands, saying "Hey Meta," and receive real-time information. The multimodal AI can translate text into different languages using the built-in camera. These capabilities were in testing for a while and are now available to everyone in the US and Canada.
Profulent’s OpenCRISPR-1 can edit the human genome
Profluent, a biotechnology company, has developed the world's first precision gene editing system using AI-generated components. They trained LLMs on a vast dataset of CRISPR-Cas proteins to generate novel gene editors that greatly expand the natural diversity of these systems. OpenCRISPR-1 performed similarly to the widely used SpCas9 gene editor regarding on-target editing activity but had a 95% reduction in off-target effects. This means OpenCRISPR-1 can edit the human genome with high precision.
The researchers further improved OpenCRISPR-1 by using AI to design compatible guide RNAs, enhancing its editing efficiency. Profluent publicly released OpenCRISPR-1 to enable broader, ethical use of this advanced gene editing technology across research, agriculture, and therapeutic applications. By using AI-generated components, they aim to lower the cost and barriers to accessing powerful genome editing capabilities.
Enjoying the weekly updates?
Refer your pals to subscribe to our newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
NVIDIA acquires Run:ai, integrates it with DGX Cloud AI Platform
NVIDIA has acquired Run:ai, an Israeli startup that simplifies AI hardware infrastructure management and optimization for developers and operations teams. The acquisition was made for an undisclosed sum, but sources suggest it was around $700 million.
Run:ai’s platform allows AI models to run parallel across various hardware environments, whether on-premises, in public clouds, or at the edge.
Nvidia plans to maintain Run:ai's products with their existing business model and will support Run:ai's product development within Nvidia's DGX Cloud AI platform. This platform offers enterprise users access to computing infrastructure and software for training AI models, including generative AI.
Snowflake enters the generative AI arena with Arctic LLM
Snowflake, the cloud computing company, has released Arctic LLM, a generative AI model for enterprise use. It’s optimized for generating database code and is available under an Apache 2.0 license.
Arctic LLM outperforms other models like DBRX and Llama3 in tasks like coding and SQL generation. Snowflake aims to address enterprise challenges with this model, including building SQL co-pilots and high-quality chatbots. This move aligns with the trend of cloud vendors offering specialized generative AI solutions for businesses
Monetizing generative AI to take time, says Zuckerberg
Meta CEO Mark Zuckerberg stated that it would take several years for Meta to make money from generative AI. The company is already profitable, but building advanced AI capabilities will be lengthy and costly. Monetization strategies include scaling business messaging, introducing ads or paid content, and offering larger AI models for a fee. However, it will take time for these efforts to yield significant profits.
Sanctuary AI unveils next-gen robots
Sanctuary AI, a company developing human-like intelligence in robots, unveiled its latest robot - Phoenix Gen 7. This comes less than a year after their previous generation robot.
The new robot boasts significant improvements in both hardware and software. It can now perform complex tasks for longer durations, learn new tasks 50 times faster than before, and have a wider range of motion with improved dexterity. The company believes this is a major step towards achieving human-like general-purpose AI in robots.
CIOs go big on AI!
A new Lenovo survey shows that CIOs are prioritizing integrating AI into their businesses alongside cybersecurity.
However, there are challenges hindering rapid AI adoption, such as:
Large portions of organizations are not prepared to integrate AI swiftly (e.g., new product lines, supply chain).
Security concerns around data privacy, attack vulnerability, and ethical AI use.
Talent shortage in machine learning, data science, and AI integration.
Difficulty demonstrating ROI of AI projects.
Resource constraints - focusing on AI may take away from sustainability efforts.
Despite the challenges, there is still a positive outlook on AI:
80% of CIOs believe AI will significantly impact their businesses.
96% of CIOs plan to increase their investments in AI.
Moderna and OpenAI partner to accelerate drug development
Biotech giant Moderna has expanded its partnership with Open AI to deploy ChatGPT enterprise to every corner of its business. The aim is to leverage AI to accelerate the development of new life-saving treatments.
Here’s the gist:
Moderna plans to launch up to 15 new mRNA products in 5 years, including vaccines and cancer treatments.
Their custom "Dose ID" GPT helps select optimal vaccine doses for clinical trials.
Moderna saw the creation of 750+ custom GPTs with 120 ChatGPT conversations per user per week.
The redesign aims for a lean 3,000-employee team to perform like 100,000 with AI force multiplication.
That's all for now!
Subscribe to The AI Edge and gain exclusive access to content enjoyed by professionals from Moody’s, Vonage, Voya, WEHI, Cox, INSEAD, and other esteemed organizations.
Thanks for reading, and see you on Monday. 😊