Samsung Debuts Home AI Robot "Ballie"

Good morning. It’s Wednesday, January 10th.

Did you know: On this day in 2007, Apple unveiled the original iPhone?

In today’s email:

  • AI Integration in Consumer Electronics & Vehicles

  • Advancements in AI & Machine Learning Models

  • AI in Business, Legal, and Government Applications

  • AI in Health, Biotech, and Science

  • 9 New AI Tools

  • Latest AI Research Papers

  • ChatGPT Writes Comics

You read. We listen. Let us know what you think by replying to this email.

Interested in reaching 47,902 smart readers like you? To become an AI Breakfast sponsor, apply here. 

Today’s trending AI news stories

AI Integration in Consumer Electronics

> At CES 2024, Lenovo launched over 40 AI-driven devices, broadening its commitment to making AI accessible to all. The lineup includes advanced AI PCs under the includes advanced AI PCs under the Yoga™, ThinkBook™, ThinkPad™, ThinkCentre™, and Legion™ brands, reforming computing experience for both consumers and businesses. The announcement also features two proof of concept products, a tablet, a software app, Motorola AI features, and accessories, enhancing Lenovo’s robust technology solutions portfolio. These innovations signify Lenovo’s ongoing journey in integrating AI into everyday technology, offering personalized, efficient, and creative computing solutions.

> Rabbit, an AI startup, has launched R1, a $199 device functioning as a versatile voice controlled universal app controller. R1’s uniqueness lies in its Large Action Model (LAM), trained using human-app interaction data, enabling it to operate existing apps without needing native support. The Rabbit R1, boasting a 2.3GHz processor, 4GB RAM, and 128GB storage, features a 2.88-inch touchscreen, a rotating camera, and a multi-functional scroll wheel. Its Rabbit OS allows it to learn and perform complex tasks, including those in applications like Photoshop. The R1, poised as a next-gen AI assistant, is expected to ship in March and is currently available for pre-order.

> Samsung’s updated Ballie robot, unveiled at CES 2024, now doubles as a projector. This AI-equipped robot can follow users around the home, project images and videos onto various surfaces, and handle smart home tasks like turning on lights and feeding pets. It can also conduct phone calls and display video calls on walls. While Samsung has showcased Ballie through a video presentation, details about its release date or pricing remain undisclosed. The robot is part of Samsung’s ongoing exploration of smart home technology and AI integration.

> Volkswagen is poised to transform the driving experience by mid-year, introducing ChatGPT as an interactive voice assistant in its vehicles. Unveiled at the CES trade fair in Las Vegas, this innovative feature will allow drivers in North America and Europe to engage in fluid dialogues with their cars, starting in the second quarter. Developed in partnership with Cerence, this technology enables cars to understand and respond to diverse commands, making tasks like adjusting the temperature or finding a restaurant as simple as speaking. This advancement by Volkswagen represents a significant stride in embedding a sophisticated AI into everyday automotive use.

Advancements in AI & Machine Learning Models

> The Allen Institute for AI’s latest creation, Unified-IO 2, might just be giving us a sneak peek into what GPT-5 could be. This model stands out for its ability to handle multitude of inputs and outputs including text, images, audio, video, and even action sequences. It’s built on a mere 7 billion parameters and was meticulously trained with a vast array of multimodal data (1 billion image-text pairings, 1 trillion text tokens, and a plethora of video, image, and 3D content.) Excelling in more than 35 diverse benchmarks, Unified-IO 2 is not just a step but a leap forward in the realm of AI, showcasing the immense potential of multimodal training in understanding and generating complex, cross-medium content.

> Mistral AI’s Mixtral 8x7B, an open-source large language model (LLM), is currently outperforming GPT-3.5 according to recent benchmarks. Detailed in their latest paper, Mixtral 8x7B surpasses similar models in language comprehension, particularly in the MMLU benchmark. Although larger models like GPT-4 still lead in some areas, Mixtral 8x7B's achievements highlight the growing capability of open-source LLMs in reaching or exceeding the performance level of GPT-3.5. | Try Mixtral 8×7B here

> Pennsylvania has become the first state to integrate ChatGPT Enterprise into its government operations, under Governor Josh Shapiro’s initiative. This pilot program, a collaboration with OpenAI, aims to enhance state employees’ efficiency but not replace them. ChatGPT will assist in tasks like drafting documents and generating code, without direct integration with citizens. The program will test OpenAI’s security measures with government-sensitive materials and act as a precursor for other states. This move coincides with OpenAI facing a lawsuit from The New York Times over copyright issues and the launch of its GPT Store.

> The European Commission is considering an antitrust investigation into Microsoft’s partnership with OpenAI, focusing on whether it complies with the EU Merger Regulation. This probe follows similar actions by the UK’s Competition and Markets Authority (CMA) and a preliminary review by the U.S. Federal Trade Commission. Microsoft’s reported $13 billion investment in OpenAI since 2019, including a $10 billion collaboration announced last year, forms the basis of this scrutiny. The partnership, which includes profit-sharing and potential stock acquisition by Microsoft, raises questions about antitrust implications, ownership stakes, and the evolving governance structure of OpenAI.

AI in Health, Biotech, and Science

> NuraLogix has introduced the Anura MagicMirror, an innovative health-monitoring device. This 21.5 inch smart mirror uses sensors and AI to analyze facial blood flow, providing insights into over 100 health parameters in just 30 seconds. It can assess various vital signs and disease risks, including blood pressure, heart rate, and risks for conditions like diabetes and cardiovascular disease. Targeted for use in clinics and health facilities, details about its launch and pricing are yet to be announced.

> Microsoft and the Pacific Northwest National Laboratory discovered a new material that could drastically reduce lithium use in batteries by up to 70%. This breakthrough, achieved in less than nine months, utilized AI to screen over 32 million potential materials, speeding up a process that could have taken decades. The material, a solid-state electrolyte, has already powered a lightbulb and shows promise for sustainable, safer, and faster-charging lithium batteries. This innovation could significantly impact the rapidly growing demand for lithium, driven by the surge in electric vehicles and other battery-powered devices.

5 new AI-powered tools from around the web offers a developer portal enhanced with AI assistant. It maps software ecosystems, unifies engineering stacks, and trains custom AI assistants, streamlining DevOps inquiries across various tools.

Stylar AI makes AI-assisted design more accessible and customizable with a user-friendly, drag-and-drop interface. It enables precise composition, layering, and style application making it ideal for graphic designers, illustrators, artists, and marketers.

Tablize is an AI-enhanced data management tool. It enables conversion of diverse data sources into dynamic dashboards and has data visualization catering to both data professionals and casual users, with features for real-time processing and intuitive dashboard creation.

Neuralhub is a pioneering platform designed for researchers, students, and engineers to design and build neural network architectures. It offers a suite of tools, libraries, and a collaborative playground for experimenting, inventing, and sharing new AI architectures.

Bland Turbo is a conversational AI capable of managing over 500,000 phone calls simultaneously, responding with human-level speed in any voice, and programmable for diverse tasks.

Sama AI is a wearable technology that acts as an AI-powered mentor, designed to improve memory and provide active support. Compatible with smartphones, it actively listens, responds with feedback, and assists in recalling conversations, all while emphasizing privacy and secure consent-based use.

Items Design offers a vast library of free AI-generated design assets, including abstract backgrounds, shapes, and real objects. Plus, each premium asset includes a unique Midjourney prompt for enhanced creativity.

Followr redefines social media management with AI, offering features like in-app image, video editing, automatic post creation, AI-powered templates, analytics, workflow automation, multi-platform support, and AI-recommended content creation., an open-source, offline chatbot, boosts productivity with customizable AI assistance and global hotkeys. Featuring an OpenAI-equivalent API server (localhost:1337), it integrates with apps and will soon launch a mobile app.

arXiv is a free online library where researchers share pre-publication papers.

The paper presents an efficient alternative to large-scale language models like ChatGPT. This study introduces 'Blending', a novel technique combining smaller models (6B/13B parameters) to achieve or even surpass the performance of much larger models (175B+ parameters). Extensive A/B testing on the Chai research platform demonstrates that these blended models significantly improve user retention and engagement. This approach offers a compelling solution to the high computational costs and memory demands associated with gigantic AI models, showcasing the power of collaborative, smaller-scale models in conversational AI.

The paper introduces a new evaluation metric for text-to-3D generative models. This approach utilizes GPT-4V to automatically generate prompts and compare 3D shapes against user-defined criteria, aligning closely with human preferences. The system can efficiently rank models using Elo ratings, offering a comprehensive assessment of text-to-3D models across various criteria. The research demonstrates that this method surpasses existing metrics in aligning with human judgment, providing a scalable and holistic way to evaluate these models.

MagicVideo-V2, developed by Bytedance Inc., is a groundbreaking multi-stage text-to-video (T2V) generation tool. It seamlessly integrates text-to-image, image-to-video, and video frame interpolation modules to produce high-aesthetic, smooth videos. The tool outperforms leading T2V systems through superior fidelity and user evaluations. Its innovative architecture involves a diffusion-based T2I model, a high-aesthetic SD1.5 I2V module, and an advanced V2V module for super-resolution, all culminating in aesthetically pleasing, high-resolution video output.

MAGNET, developed by Meta’s FAIR Team and others, is an innovative audio generation method using a non-autoregressive transformer. It creates complex audio sequences from textual descriptions, significantly faster than autoregressive methods. MAGNET’s single-stage transformer model predicts spans of masked audio tokens, enhancing audio quality through a novel rescoring method with an external pre-trained model. It also explores a hybrid approach, combining autoregressive and non-autoregressive models. This method excels in text-to-music and text-to-audio generation, achieving comparable results to baselines with reduced latency, confirmed by extensive empirical evaluations including human studies.

In the quest for efficient and powerful conversational AI, the study "Blending Is All You Need" challenges the norm of scaling up model size, notably against the backdrop of giants like ChatGPT with its 175B+ parameters. The researchers from the University of Cambridge and University College London propose "Blended," an approach that synergizes multiple smaller models (6B/13B parameters) to potentially surpass or match the performance of larger counterparts like ChatGPT. This hypothesis was tested through extensive A/B testing on the Chai research platform, where Blended demonstrated higher user retention and engagement, despite being significantly more resource-efficient. Blended leverages the strengths of individual models in a collaborative manner, creating diverse and engaging conversations. The findings reveal that a collective of smaller, efficiently run models can effectively compete with a single, larger model in terms of user experience, marking a shift in the approach to developing conversational AIs.

ChatGPT Writes Comics

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.