Anthropic launches most intelligent AI yet

Research: OpenAI pivots to new training method

Happy Friday!

We’re capping off a relatively quiet week in AI with huge news from some of the biggest names in the industry. Let’s get straight to it.     

In today’s Daily Update:

  • 🗞️ Anthropic launches Claude 3.5 Sonnet   

  • 🎂 OpenAI turns to new AI training approach     

  • 📸 Nvidia and Dell team up to build xAI supercomputer                

  • 🚨 AI Roundup: Four quick hits

Read time: 2.5 minutes

TOP STORY

🗞️ Anthropic launches Claude 3.5 Sonnet

Source: Anthropic

Anthropic just launched its most powerful LLM to date. The company says Claude 3.5 Sonnet raises the industry bar for intelligence across a wide range of tasks.

What you should know:

  • Claude 3.5 Sonnet sets new industry benchmarks for graduate-level reasoning, undergraduate-level knowledge and coding proficiency. 

  • The model features improved visual reasoning, allowing it to interpret charts and graphs, transcribe text from images and solve problems based on visual information. 

  • Anthropic says the new model is twice as fast as its predecessor Claude 3 Opus without increasing usage costs. 

  • Claude 3.5 Sonnet appears to outperform OpenAI’s GPT-4o across most major benchmark tests. The model is available for free at claude.ai

Why it matters: Backed by Amazon and Google, Anthropic continues to develop competitive models despite being measurably smaller than OpenAI. Future Claude iterations could potentially challenge ChatGPT's market position, especially if Anthropic expands their multimodal capabilities in areas like image generation and audio processing.

(Side note: I tested Claude 3.5 Sonnet last night and it had me reconsidering my ChatGPT Plus subscription. Definitely recommend checking it out.)

RESEARCH INSIGHT

🎂 OpenAI turns to new AI training approach

DALL-E 3

OpenAI researchers have introduced a new AI method called “consistency models,” which generate high-quality images much faster than current diffusion models. 

The difference explained through cake:

  • Diffusion: Imagine you’re trying to create the perfect recipe for a cake. Diffusion models would start with a perfect cake and gradually add random ingredients, effectively messing up the cake through a series of small steps. The model then learns how to reverse this process by removing the random ingredients one by one. To make a new cake, it would start with random ingredients and slowly refine them into a perfect cake.

  • Consistency: These models learn how to jump from random ingredients to the perfect cake in one big step. Alternatively, this process can be broken up into a few sizable steps if needed. The goal is to achieve the same result as diffusion models but much faster. 

Why it matters: Diffusion models can make very good cakes but they do so very slowly. This report focuses exclusively on image generation and editing, but the consistency model approach could potentially be applied to train AI for other tasks. This breakthrough brings us one step closer to AI that can produce responses at humanlike speeds. 

BUSINESS SPOTLIGHT

📸 Nvidia and Dell team up to build xAI supercomputer

“Grok” by DALL-E 3

Nvidia and Dell are joining forces to build an AI factory designed to improve Grok, the flagship model developed by Elon Musk’s xAI. 

The details:

  •  xAI recently announced plans to construct the world’s most powerful supercomputer (coined the “Gigafactory of Compute”) by fall 2025. 

  • The project will utilize up to 100,000 Nvidia GPUs, which could make it 4x larger than the strongest existing GPU clusters.

  • Musk says Dell will assemble half of the server racks for the Gigafactory of Compute.

  • San Francisco-based Supermicro is also set to collaborate on the factory.

Why it matters: xAI is beginning to establish itself as a legitimate competitor to OpenAI and Anthropic. Funding and infrastructure are at the core of developing AI, and Musk is building a GPU cluster that will surpass his rivals’ in terms of pure size and capabilities.

MORE TRENDING NEWS

🚨 AI Roundup: Four quick hits

DALL-E 3

  • Back to the beginning: This is how Nvidia became an AI giant.

  • Booed off stage: Comedians say AI is effectively useless at writing comedy.

  • Clone creation: HeyGen raises $60 million for its virtual avatar platform. 

  • Level up: Apple Developer Academy launches AI training for students and alumni. 

THAT’S ALL FOR TODAY

Want to continue the conversation? Connect with me on LinkedIn and I’m happy to discuss any of today’s news. Thanks for reading The Daily Update!

(P.S. If you want to share this newsletter with a friend or colleague you can find it here.)