🤩Meta's Best Open-Source AI Yet

Meta releases Llama 3.1—its biggest, best open-source AI model yet

Key Points 

• Llama 3.1 has a context length of 128k, can chat in eight languages, write better computer code, and solve more complex math problems.

• CEO Mark Zuckerberg strongly believes that by the end of the year, Meta AI will be the most widely used assistant, easily surpassing ChatGPT.

• In the US, you can access the new model via Meta AI with your Facebook or Instagram account, outside the US, you can use it for free on HuggingChat without signing up.

 ♨️News - Back in April, Meta hinted it was working on something big: an open-source AI model that could compete with the best from companies like OpenAI. Now, that model is here. 

Meta just released Llama 3.1, their largest open-source AI model yet, and they say it outperforms GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks. CEO Mark Zuckerberg strongly believes that by the end of the year, Meta AI will be the most widely used assistant, easily surpassing ChatGPT.

🤩What's so special about it? Llama 3.1 is much more advanced than the smaller Llama 3 models that were released a few months back. This new model has a context length of 128k, can chat in eight languages, write better computer code, and solve more complex math problems. According to Meta’s blog posts and research paper, the largest version has 405 billion parameters and was trained using over 16,000 of Nvidia’s high-end H100 GPUs.

👨🏻‍💻So then why open-source it? According to a letter from Mark Zuckerberg on Meta’s blog, he believes open-source AI models will eventually outpace proprietary ones, much like how Linux became the dominant open-source operating system.

Zuckerberg compares Meta’s investment in open-source AI to its earlier Open Compute Project, which saved billions by collaborating with companies like HP to improve and standardize data center designs. He expects a similar outcome with AI, predicting that Llama 3.1 will mark a turning point where most developers start using open-source models.

Meta is partnering with over two dozen companies, including Microsoft, Amazon, Google, Nvidia, and Databricks, to help developers deploy their own versions. Meta says that running Llama 3.1 costs about half as much as OpenAI’s GPT-4o. They’re also releasing the model weights so companies can train it on their own data and customize it as needed.

🤓Not a developer but still want to try Llama 3.1? If you’re in the US, you can use it through Meta AI by signing in with your Facebook or Instagram account. Just enter your prompt and the model will respond. 

If you're outside the US, you can still use the model on HuggingChat. It’s free, and you can chat with the model without signing up here.

Adobe launches new Firefly AI tools for Illustrator and Photoshop

Key Points 

• Illustrator: New "Generative Shape Fill" tool adds details and textures to shapes using text prompts or style references, with adjustable detail and editing options.

• Photoshop: Firefly’s text-to-image generator is now generally available for creating AI images. The Enhance Detail feature can now sharpen large images, and the Selection Brush tool is also generally available for easier object selection.

🎨News - On Tuesday, Adobe launched new Firefly tools for Photoshop and Illustrator, giving graphic designers even more ways to use their in-house AI models.

👨🏻‍🎨Here’s what’s new - Illustrator now includes a tool called “Generative Shape Fill.” It lets users add details and textures to shapes using text prompts or style references. You can adjust the detail with sliders and then edit the results, which is a step up from typical AI generators. This feature uses an updated Firefly Vector model, also launching today. The updated model also enhances the Text to Pattern feature for creating scalable vector patterns and the Style Reference tool for mimicking existing styles.

For Photoshop, Adobe is now making Firefly’s text-to-image generator generally available in the Photoshop desktop and web apps. You can create AI-generated images by pressing “Generate Image” and describing what you want. This feature, powered by Adobe Firefly’s Image 3 Foundation model, was previously in beta. Additionally, the Enhance Detail feature for Generative Fill now sharpens and adds more detail to large images; and the Selection Brush tool is also being made generally available, making it easier to select and isolate specific objects from the canvas by painting over them.

🖌️See also - Illustrator has also introduced several new tools that don’t use AI. You can find more details about these updates on Adobe’s blog.