Meta Releases World’s Largest Open-source LLM
Plus: Mistral’s rival to Llama 3.1, OpenAI’s SearchGPT, foundation models powering Apple Intelligence, and more.
Hello Engineering Leaders and AI Enthusiasts!
This newsletter brings you the latest AI updates in just 4 minutes! Dive in for a quick summary of everything important that happened in AI over the last week.
And a huge shoutout to our amazing readers. We appreciate you😊
In today’s edition:
🤖 Meta released world’s largest open-source LLM to date
🚀 Mistral AI released its Llama 3.1 rival, Mistral Large 2
🏛️ US lawmakers are requesting OpenAI for government access
🥈 DeepMind’s new AI is a silver medalist in the IMO math Olympiad
🔍 OpenAI announced SearchGPT, an AI-powered search engine
🧠 Apple revealed AI models powering Apple Intelligence
📚 Knowledge Nugget: The Missing Link between AI and APIs by
Let’s go!
Meta released largest open-source LLM ever
On July 23rd, Meta officially released the biggest version of its open-source LLM, Llama, a 405 billion-parameter version called Llama-3.1. It also released Llama 3.1 70B and 8B models.
Llama 3.1’s context window has been expanded to 128,000 tokens, meaning users can feed it as much text as in a 400-page novel. It will be multilingual and support English, Portuguese, Spanish, Italian, German, French, Hindi, and Thai.
The 405B model is competitive with leading foundation models across a range of tasks, including GPT-4, GPT-4o, and Claude 3.5 Sonnet. The smaller models also performed similarly.
Users can access Llama 3.1 through AWS, Nvidia, Groq, Dell, Databricks, Microsoft Azure, Google Cloud, and other model libraries. Llama 3.1 405B will also be available on WhatsApp and Meta AI.
Why does it matter?
The move directly challenges industry leaders like OpenAI and Anthropic, particularly OpenAI’s market-leading position. It also underscores Meta’s commitment to open-source development, marking a major escalation in the AI competition.
Mistral AI released its Llama 3.1 rival
Mistral AI has announced the next generation of its flagship open-source model with 123 billion parameters, Mistral Large 2. Compared to its predecessor, the model is significantly more capable in code generation, mathematics, and reasoning. It also provides much stronger multilingual support and advanced function-calling capabilities.
However, the model is only licensed as “open” for non-commercial research uses, including open weights, allowing third parties to fine-tune it to their liking. Those seeking to use it for commercial/enterprise-grade applications will need to obtain a separate license and usage agreement from Mistral.
Why does it matter?
Following Meta’s launch of Llama 3.1 as a highly competitive alternative to leading closed-source “frontier” models, the French AI startup entered the fray. The AI race is picking up pace like never before.
US lawmakers request OpenAI for government access
Five U.S. Senators sent a letter to OpenAI CEO Sam Altman, demanding details about the company's safety standards and employment practices.
Perhaps the most significant portion of the letter was item 9: "Will OpenAI commit to making its next foundation model available to U.S. Government agencies for pre-deployment testing, review, analysis, and assessment?”
The letter outlined 11 additional points to be addressed, including OpenAI’s commitment to dedicating 20% of its computing power to fuel safety research and protocols to prevent malicious actors or foreign adversaries from stealing OpenAI’s products or IP.
Why does it matter?
Regulatory scrutiny is nothing new for OpenAI and the broader AI sector. However, now OpenAI is facing heightened scrutiny, and following developments could drive stringent government oversight and set new industry standards.
DeepMind’s new AI is a silver medalist at IMO’24
Google DeepMind presented AlphaProof, a new reinforcement-learning based system for formal math reasoning, and AlphaGeometry 2, an improved version of its geometry-solving system.
Together, these systems solved four out of six problems from this year’s International Mathematical Olympiad (IMO), achieving the same level as a silver medalist for the first time. Here’s a graph showing the AI system's performance relative to human competitors at IMO 2024.
Why does it matter?
Solving complex math problems in step-by-step proofs has been a grand challenge for AI. Breakthroughs like these demonstrate AI’s growing ability to match top human minds, with far-reaching implications across various fields.
OpenAI announced an AI-powered search engine
OpenAI is testing SearchGPT, a prototype combining the strength of its AI models with information from the web. It will quickly and directly respond to your questions with up-to-date information while providing clear links to relevant sources. You’ll also be able to ask follow-up questions.
It is launching to a small group of users and publishers to get feedback. While this prototype is temporary, OpenAI plans to integrate the best of its features directly into ChatGPT in the future.
Why does it matter?
This directly challenges Google’s dominance in the online search market. It also signals a significant escalation in AI search wars, which are already reshaping how users find and interact with information on the web.
Apple reveals the AI powering Apple Intelligence
Apple published a research paper describing two new foundation language models that form the backbone of Apple Intelligence, its new AI system.
AFM-on-device (AFM stands for Apple Foundation Model), a ∼3 billion parameter language 1 model, and
AFM-server, a larger server-based language model
The models are designed to be fast and run efficiently on iPhone, iPad, and Mac as well as on Apple silicon servers via Private Cloud Compute. They are part of a larger family of generative models created by Apple to support users and developers.
Why does it matter?
Apple Intelligence is designed with Apple’s core values at every step and a foundation of industry-lead privacy protection, showing Apple’s commitment to providing secure, powerful, personalized AI experiences.
Enjoying the latest AI 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.
Knowledge Nugget: The Missing Link between AI and APIs
In this insightful article,
discusses the current limitations of AI systems like ChatGPT in connecting to and using external APIs. He argues that despite improvements in conversational abilities, AI still can't perform tasks like booking tickets directly.The article critiques the current approach of custom GPTs, arguing that AI should be able to automatically discover and use relevant APIs based on user intent rather than requiring users to find specific "AI apps" for tasks.
Why does it matter?
The ability of AI to seamlessly integrate with external services and APIs would represent a significant leap forward in making AI systems truly useful for everyday tasks. The article urges a different approach to AI development to address this gap.
What Else Is Happening❗
🔥OpenAI intensified the AI arms race by announcing free fine-tuning for its GPT-4o Mini model, just hours after Meta launched its open-source Llama 3.1 model.
🎥Stability AI released Stable Video 4D, its first video-to-video AI model that turns a single object video into multiple novel-view videos with eight different angles/views.
📉A new study found indiscriminate use of AI-generated data in training leads to irreversible defects, termed "model collapse," where the models plateau and become incoherent.
🔍Bing released its answer to Google’s AI-powered search, Bing generative search. It is currently available to a small percentage of users for preview.
🌍Kling AI has gone global with an International Version 1.0 to take on OpenAI’s yet-to-be-released video generator, Sora. It is now accessible to all at KlingAI.com, where registration requires only an email address.
🌟Google introduced 1.5 Flash in the unpaid version of Gemini for faster and better responses. It also introduced a new feature to further address hallucinations and expanded Gemini for Teens and mobile apps.
🚫X now automatically activates a setting that allows it to train its Grok AI on user data, including posts, user interactions, inputs, and results. Find out how you can switch it off!
🤖Meta launched AI Studio, a platform built on Llama 3.1 that lets anyone create share, and discover AI characters and allows creators to build an AI as an extension of themselves to reach more fans.
🚀Amazon has reportedly unveiled a new AI chip, boasting 40-50% higher performance than NVIDIA's at half the cost, aiming to reduce reliance on expensive external chips.
🤗Hugging Face is offering developers an inference-as-a-service powered by Nvidia NIM microservices. It will improve token efficiency by up to 5x with popular AI models.
New to the newsletter?
The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From machine learning to ChatGPT to generative AI and large language models, we break down the latest AI developments and how you can apply them in your work.
Thanks for reading, and see you next week! 😊