Ahmad Al-Dahle’s Post

I am thrilled to welcome Russ Salakhutdinov to AI at Meta as our newest VP of Research in GenAI! I’ve known Russ for a number of years, including our time together at Apple and couldn’t be more excited for what’s ahead as he helps us deliver on some of our big bets for AI agents.

We just announced really impressive new work on generative 3D models from the GenAI Research team at Meta. Meta 3D Gen is a new system for end-end generation of 3D assets from text that’s capable of outputting high-resolution textures and materials in less than one minute — up to 10x faster than existing approaches and setting a new state-of-the-art. Awesome work from Natalia Neverova and team, I’m excited for the possibilities this will open up with the new bar for quality and performance.

We built Meta AI to be one of the most advanced AI assistants in the world and it’s really cool to see the unique ways people are now using it. This week, our team started rolling out a consumer campaign showcasing the different ways Meta AI can have an impact in people’s daily lives — plus a small preview of some things we’re excited to ship soon. I’m proud of all of the work from the GenAI team that went into these experiences, and I’m even more excited about the new capabilities coming to Meta AI in the coming months.

We're hosting the first ever Llama 3 Hackathon together with Cerebral Valley and SHACK15 — I couldn’t be more excited to join the community as they build with our newest models over two days in San Francisco. Many of our Llama team members will be onsite to discuss their work and support hackers as they build. 🏆 Prizes include $10K+ in cash + credits to kickstart Llama 3 projects. Register to join us from May 11-12: https://lnkd.in/g2_ayKWu

What a week since we released Meta Llama 3! It’s been really incredible to see the way the community has responded to these new models. We’re already seeing inspiring use cases out in the wild, like Yale’s School of Medicine + EPFL School of Computer and Communication Sciences, who fine-tuned a version of the Llama 3 for medicine that’s already best-in-class vs comparable models in benchmarks like MedQA and MedMCQA. We shared more of our reflections on this week in this blog post: https://lnkd.in/geJGEDE4 A few more of my personal highlights: 🏆 Llama 3 70B is now the highest ranking open model on the LMSYS Chatbot Arena leaderboard — and tied for #1 in English. 📈 We’ve already seen Llama 3 models downloaded over 1.2M times — roughly 3x faster than Llama 2! 🤗 The community has been hard at work, there are already 600+ derivative models built on Llama 3 released on @Hugging Face. ⭐ The new meta-llama repo on GitHub already has over 17K stars. The rapid pace of innovation on our Llama models is wild to watch. We’re just getting started with Llama 3 and I can’t wait to see what the coming weeks and months hold as we release more models and watch the community do exactly what the community does best. 

If you haven’t tried imagine flash, highly recommended trying it at meta.ai. Just log in and type “imagine <your prompt>” and watch the model generate images as you type, it’s magic. Excellent work by Roshan Sumbaly , Péter Vajda and Ali K. Thabet ‘s groups

One of the coolest things about Meta AI is how we’ve taken the experience to extend beyond just our apps and onto our hardware devices. With the latest update, rolling out today on Ray-Ban Meta smart glasses, we’re taking things even further with multimodal AI that helps you understand and interact with the world around you in new ways. It’s a pretty awesome experience that you just have to try for yourself — it introduces new ways Meta AI can be helpful in your day-to-day life. Our team is hard at work on some exciting things for multimodal AI and I’m excited to share more in the coming months around how we’ll bring these capabilities to life on Meta AI. 

Lots to talk about this week! In addition to launching Llama 3 yesterday, we also released the latest version of Meta AI, with a number of major improvements. 🦙 Meta AI is now powered by Llama 3.  🌏 We expanded to 13 new countries. ❤️ Made Meta AI more prominent across all of our apps. 🌐 Launched our meta.ai website. 🔎 Added search powered by Google! Meta AI is now smarter, faster and more fun than ever before. Try it out across our apps or at meta.ai! Our newest version of Meta AI also comes with a significant upgrade in our image generation capabilities. Now on Meta AI you can imagine new images faster, with higher quality and even animate them — all through a simple text prompt. It’s awesome. You can also create images in real time on WhatsApp and meta.ai, that will update with every letter you type. Give it a try and let me know what you think.

It’s finally here! I’m thrilled to introduce you to Llama 3, our latest generation of models that is setting a new standard for state-of-the art performance and efficiency for openly available LLMs. Key highlights: • 8B and 70B parameter openly available pre-trained and fine-tuned models • Trained on more than 15T tokens, 7x+ larger than Llama 2's dataset! • State-of-the-art performance across industry benchmarks • New capabilities, including enhanced reasoning and coding • 3x more efficient training than Llama 2 • New trust and safety tools with Llama Guard 2, Code Shield, and CyberSec Eval 2. • Training underway on a leading-edge 400B+ parameter model  We’re also releasing a new version of Meta AI powered by Llama 3. We’re making it available in 13 new countries, integrating enhanced search capabilities and adding new ways to experience it on the web at meta.ai. Our latest updates to image generation on Meta AI is state of the art, allowing you to create, animate and share stunning images with a simple text prompt in seconds. We think Meta AI is the world’s most intelligent AI assistant and we can't wait for you to try it and tell us what you think. Read more about it here: https://lnkd.in/eB5yzi8R