- AI Breakfast
- Posts
- Could Meta's Llama 3 surpass GPT-4 this year?
Could Meta's Llama 3 surpass GPT-4 this year?
Plus, our book is 70% off today only....
AI Breakfast
August 28, 2023
Good morning. It’s Monday, August 28th.
Did you know: Google’s Gemini may surpass GPT-4 training by 5x this year?
In today’s email:
AI Technology & Innovations
Autonomous & AI-Powered Vehicles
AI in Defense & Warfare
AI Models & Competition
AI in Legal & Governance
AI in Biotech & Health
5 New AI Tools
Latest AI Research Papers
You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.
Today’s edition is brought to you by:
Our book, Decoding AI: A Non-Technical Explanation of Artificial Intelligence is on sale for just $2.99 today only!
(with a 100% money-back guarantee)
Decoding AI breaks down the complexities of AI into digestible concepts, walking you through its history, evolution, and real-world applications.
We'll introduce you to the key players in the AI field, as well as explain the underlying algorithms, data, and machine learning concepts that power AI systems. You'll gain a deeper understanding of deep learning, neural networks, and reinforcement learning, and we'll explore various types of AI, from rule-based systems to probabilistic networks and beyond.
The goal was to make this book an approachable discovery of how AI works.
It discusses a wide range of applications AI has in areas like natural language processing, computer vision, robotics, and predictive analytics. It also delves into the regulatory landscape and policy issues surrounding AI, as well as the potential future developments in AI, such as its applications in healthcare, education, transportation, and even space exploration.
You'll also learn the difference between narrow AI and Artificial General Intelligence (AGI), and how to get started with using AI through tips and resources.
Price goes back to $9.99 after today’s sale. 100% money-back guarantee if not satisfied.
Today’s trending AI news stories
AI Technology & Innovations
Machine-learning system based on light could yield more powerful, efficient large language models MIT researchers have developed a machine-learning system that utilizes light for computations, potentially enabling significantly more powerful machine-learning models with vastly improved energy efficiency. The system, detailed in the journal Nature Photonics, uses micron-scale lasers to perform computation based on light movement rather than electrons. It demonstrates a more than 100-fold improvement in energy efficiency and a 25-fold improvement in compute density compared to current state-of-the-art systems. The researchers believe this approach could pave the way for large-scale optoelectronic processors that could be used for machine-learning tasks in data centers and decentralized edge devices.
Google DeepMind's new chess engine beats its famous AlphaZero Google DeepMind develops a multi-agent chess engine, AZdb, by combining various AlphaZero agents into a “league” to enhance chess capabilities and improve generalization. AZdb’s agents employ “behavioral diversity” and “response diversity” techniques to develop unique playing styles and better adapt to different opponents and unseen positions. Testing showed that AZdb solved twice as many challenging chess puzzles compared to AlphaZero, resulting in a 50 ELO rating increase. This approach suggests that incorporating human-like creativity and diversity can improve AI’s ability to generalize.
AI revolution in video games has industry players treading warily Artificial Intelligence takes center stage at Gamescom, a major video game industry fair, with AI being used for various aspects of game development, including generating storylines, coding games, and creating animations. However, concerns arise within the industry about potential job redundancies and the impact of artistic creativity. While AI offers benefits such as producing illustrations from text and enhancing gameplay realism, there are worries about the displacement of certain roles, such as concept artists. The gaming community is exploring AI-driven advancements while navigating the ethical and creative implications.
Autonomous & AI-Powered Vehicles
Elon Musk takes FSD Beta V12 on 45-minute rush hour demo while musing over stop sign regulations The FSD V12 showcases improved smoothness and intuitive handling of complex maneuvers like unprotected left turns. The AI-driven software processes video data to interpret road conditions, eliminating the need for individually coded scenarios. Despite regulatory challenges, the FSD V12 shows promise in real-world applications, marking a significant step forward in autonomous driving technology.
AI in Defense & Warfare
Air Force Wants $5.8 Billion to Build AI-Powered Valkyrie Aircraft The US Air Force seeks a budget of $5.8 billion to develop AI-powered XQ-58A Valkyrie aircraft for autonomous missions, raising concerns over the ethical implications of using technology for lethal purposes. These unmanned aircraft are designed to serve as robotic wingmen to human pilots, suitable for scenarios where human involvement might be risky, such as suicide missions. Advocates argue that outsourcing killing to machines blurs moral boundaries, while proponents of AI weapons advancement worry about the potential for rapid conflict escalation and the creation of weapons of mass destruction.
AI Models & Competition
Meta plans to take on GPT-4 with a rumored Llama 3, which is still free: Meta is reportedly developing Llama 3, aiming to compete with GPT-4 in the AI language model landscape. This potential move was revealed by OpenAI engineer Jason Wei, who overheard discussions at a Meta event. Llama 3 is rumored to match GPT-4’s performance level and maintain a free licensing model similar to previous Llama versions. Although Wei is a credible source, the plans might change, and no official release date for Llama 3 has been confirmed. Meta’s AI strategy seems to challenge OpenAI’s dominance while leveraging community development for its Llama models.
China leaps forward in the A.I. arms race as Alibaba releases a new chatbot that can ‘read’ images: Alibaba has introduced two new AI models, Qwen-VL and Qwen-VL-Chat, which focus on “reading” images and offer capabilities like providing directions via scanned street signs, solving math equations from photos, and creating narratives based on multiple pictures. This release marks China’s advancements in the AI field and puts ALibaba in competition with models like ChatGPT and Google Bard. The company’s focus on image-scanning technology aims to aid visually impaired individuals with tasks like shopping. These models will be accessible on Alibaba Cloud’s Modelscope platform and HuggingFace’s library of AI models.
AI in Legal & Governance
Courts to trial AI to draft rulings Taiwan’s Judicial Yuan and Chunghwa Telecom are set to launch a trial AI program for drafting court rulings. Developed jointly, teh AI program aims to streamline court processes and reduce clerks’ workloads. The program utilizes the MT5 large language model and has been trained on rulings and legal terminology from 1996 to 2021. Judges have reportedly been satisfied with preliminary results, which suggest the program could assist in generating ruling notices for cases such as driving under the influence (DUI) and aiding and abetting fraud. The system’s potential to decrease drafting time and improve efficiency is being explored.
AI in Biotech & Health
Ex-Meta Researchers Have Raised $40 Million From Lux Capital For An AI Biotech Startup Former Meta researchers have founded a startup named EvolutionaryScale, raising at least $40 million for their venture. Led by Alexander Rives, the team developed an AI model for biology similar to GPT-4 or Google’s Bard, but trained on protein molecule data to predict protein structures. The startup aims to use its AI model to assist in drug development, disease treatment, and biotech applications. Lux Capital led the funding round, valuing EvolutionaryScale at $200 million. The team’s expertise and focus on scaling AI in the biology field set them apart in an industry where such efforts are scarce.
Groundbreaking AI-Method Finds a Way to People’s Hearts Researchers at Osaka Metropolitan University have developed an AI model that accurately categorizes cardiac functions and identifies valvular heart diseases using chest X-rays. The AI model’s accuracy in classifying six types of valvular heart disease ranges from 0.83 to 0.92 (AUC), making it a potentially useful tool in diagnosing heart conditions. The AI model could be especially valuable in settings with a shortage of specialized technicians, as it can supplement traditional echocardiography for diagnosing heart diseases. The study showcases the potential of AI in improving medical diagnostics and patient outcomes.
5 new AI-powered tools from around the web
Storyboarder.ai, developed by FYNAL, revolutionizes professional storyboarding with AI-generated visuals. Its table-based interface streamlines shot list creation and editing, while advanced algorithms ensure high-quality shot images.
Hologram Looking Glass Blocks platform is a holographic sharing platform for 3D creators. Easily transform 2D images into holograms, embed them online, and view on any device. A hub for 3D creators to share and explore holograms ushering in a new era of visual sharing.
DigitbiteAI transforms businesses with its AI suite, featuring content and image generation, Ai chat, text-to-speech, transcription, and custom tools fueled by OpenAI’s GPT3.5 and GPT-4. Elevate content creation, customer engagement, accessibility, and efficiency using cutting-edge AI-powered solutions.
Ideogram empowers users to graphically represent data using diagrams and charts. With an intuitive interface, it simplifies visualization of data points, relationships, and trends.
AI Analytics by skills.ai accelerates data analysis with AI-driven code generation and compelling visualization. Simplifying complex datasets, it automates code creation and debugging. Craft insightful presentations and transform raw data into actionable insights easily.
arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.
The paper presents a framework for educators to leverage Generative AI (GenAI) in creating course content. As GenAI gains prominence among students for academic assignments, the authors emphasize its potential for rapid and diverse content creations. The paper introduces the GenAI Content Generation Framework, guiding educators in utilizing GenAI effectively for designing university-level course material. The framework offers practical strategies, aligning content creation with educational goals. The authors discuss the benefits and challenges of incorporating GenAI emphasizing the transformative impact it can have on education. The framework aims to empower educators and stakeholders to harness GenAI’s potential while navigating its complexities.
In this paper, the authors propose a deep learning-based method for further compressing JPEG images losslessly. They introduce a Multi-Level Parallel Conditional Modeling (ML-PCM) architecture that enables parallel decoding in different granularities. This architecture processes luma and chroma components independently, and they design efficient context models for both components. The proposed method achieves better compression ratio and lower latency compared to previous state-of-the-art methods. The approach is also extended to act as a lossless JPEG codec with advantages at higher bit rates. Experiments demonstrate its effectiveness and efficiency, making it suitable for real-world applications.
The paper presents a framework for educators to leverage Generative AI (GenAI) in creating course content. As GenAI gains prominence among students for academic assignments, the authors emphasize its potential for rapid and diverse content creations. The paper introduces the GenAI Content Generation Framework, guiding educators in utilizing GenAI effectively for designing university-level course material. The framework offers practical strategies, aligning content creation with educational goals. The authors discuss the benefits and challenges of incorporating GenAI emphasizing the transformative impact it can have on education. The framework aims to empower educators and stakeholders to harness GenAI’s potential while navigating its complexities.
The paper presents POLCA, addressing the power efficiency challenge posed by the increasing demand for large language models (LLMs) in datacenter GPUs. The authors analyze power consumption patterns in LLM clusters during prompt and token phases, highlighting opportunities for power oversubscription. They propose the POLCA framework, enabling safe and efficient oversubscription, and validate it through simulations, demonstrating up to 30% more servers in existing GPU clusters for inference with minimal performance loss. POLCA offers cost-efficient scaling while reducing data center carbon footprint, meeting LLM workload demands.
The paper introduces the Visual Instruction Generation Correction (VIGC) framework to autonomously generate high-quality image-text instruction fine-tuning datasets for multimodal models. Existing approaches rely on language-only models and manual annotations, limiting scalability and diversity. VICG utilizes multimodal models, VIG and VIC, to generate diverse question-answer pairs and refine generated answers iteratively to mitigate hallucination. Different instruction types are employed, including conversation, detailed description, and complex reasoning. Extensive experiments on LLMs show that VIGC data enhances model performance on various benchmarks. This work provides a self-guided, model-driven approach for generating high-quality instruction data for vision-language tasks, addressing limitations of existing methods.
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.