Watch out for Anthropic...

Claude 3 Dethrones GPT-4

Anthropic’s Claude 3 just managed the unthinkable: The chatbot dethroned GPT-4 from its long-held position as the world’s best LLM. But that’s only part of Anthropic’s wild week…

What happened: On Tuesday, Anthropic’s Claude 3 Opus model hit first place on the LMSYS Chatbot Arena Leaderboard (a crowdsourced open platform for evaluating AI models). Ever since the leaderboard launched in May last year, GPT-4 has always been at the top of the chart…until now. 

What does that mean? Chatbot Arena lets users chat with two anonymous AI chatbots and pick which one produces better outputs. The platform aggregates the choices from its thousands of users to rank the best LLMs on a leaderboard that updates over time.

Comparing performance: Users have been loving Claude 3 Opus for advanced tasks like investment analysis, advanced prompt engineering, and MRI scan diagnoses. But Haiku—the smallest of the Claude 3 models—is also making a splash:

• Haiku just beat an earlier version of GPT-4 on the same leaderboard. 

• That’s especially impressive because Haiku is significantly smaller and cheaper than GPT-4 (we’re talking ~120x cheaper for input tokens).

And that’s not all for Anthropic: The company just secured a $2.75 billion investment from Amazon, adding onto its initial $1.25 billion check. The craziest part? That’s Amazon’s largest outside investment in 30+ years.

• This is part of a deal by which Amazon promised to fund Anthropic with up to $4 billion by the end of March in return for a minority stake. With just one week to go, Amazon maxed out its investment.

• The substantial financial backing makes Anthropic a serious long-term rival for OpenAI. 

Image Generation Just Got 30x Faster

If image generators haven’t impressed you yet, they’re about to. A new method developed by MIT researchers turbocharges the most popular models—making them 30x faster.

Some background: Diffusion models (i.e. the models powering image generators like Stable Diffusion and DALL-E 3) work by starting with a random pattern of noise and gradually shaping it into a detailed image. It’s a notoriously slow process.

Enter Distribution Matching Distillation (DMD), which simplifies the multi-step process of diffusion models into a single step: 

• With DMD, researchers first train a high-quality but slow diffusion model called the "teacher" model.

• Then, they train a second faster "student" model to mimic the teacher's behavior at each step of the image generation process.

• In a process called “distillation,” the student observes all the steps diffusion models take and condenses them into one. 

In tests, DMD slashed the image-generation time from 2.59 seconds (using Stable Diffusion v1.5) to 90 milliseconds—almost 30x faster while maintaining or improving upon the image quality. This also dramatically reduced the computing costs of image generation.

Be the First to Try the Future of AI Video

LTX Studio is a revolutionary AI storytelling platform that takes your ideas from concept to full video with ease. Just provide a prompt, and LTX generates a full visual narrative—complete with characters, scenes, and styles. Then, effortlessly bring your story to life:

• Edit characters, voiceovers, music, and visuals

• Adjust scene settings like locations and lighting

• Experiment with styles from anime to claymation

• Generate custom shot types, SFX, and camera motion

• Export as video or zip file with all assets

Want to try it for yourself? LTX Studio has started letting people in from the waitlist! Sign up for the waitlist and soon, you’ll be among the first to experience this game-changing technology.

Adobe Brings AI to Marketing

Adobe held its biannual Summit event on Tuesday in Las Vegas—and its AI announcements might just change the future of marketing.

A new partnership with Microsoft. By connecting Adobe Experience Cloud features with Microsoft Copilot, the companies aim to become a one-stop shop for generative AI-powered marketing workflows. Some features:

• Create AI-generated campaign briefs and presentations in Microsoft 365 with marketing insights from Adobe.

• Use Adobe Firefly’s generative AI to create imagery within Microsoft apps.

• Pull campaign insights from Adobe into Microsoft apps (e.g. ask Copilot for the status of a campaign while drafting an email in Outlook).

A new branding tool, GenStudio. This product combines multiple marketing tools—including asset generation, AI-powered content management, campaign tracking, and performance insights—into one platform. Users can train GenStudio on their business’ branding, product descriptions, and customer personas. 

And Firefly Services. That’s Adobe’s new suite of generative AIs and APIs (including Photoshop) that helps teams create professional marketing and product images. It automates tasks like background removal and image cropping, and even generates product images from text prompts.

• OpenAI is kicking off a pilot program for GPT monetization.

• Intel, Qualcomm, Google, and other tech giants are reportedly joining forces to compete with Nvidia.

• Chinese tech giant Baidu is in talks about providing the AI backend for Apple products in China.

• Open Interpreter’s 01 Lite is the latest AI hardware device making waves.

• OpenAI is planning new features for GPT-4 and DALL-E 3.

• All X Premium subscribers users are getting access to Grok. 

More important AI news: Dive deeper into this week’s hottest AI news stories (because yes, there are even more) in my latest YouTube video:

Awesome tools: Here are 20 AI tools that are true game-changers for content creation:

AI music has come a long way: This video walks through and demos my two favorite AI music tools: