• Starter AI
  • Posts
  • Databricks’ DBRX, Claude 3 takes the lead, MLCommons’ new benchmark

Databricks’ DBRX, Claude 3 takes the lead, MLCommons’ new benchmark

GPT-4 has been dethroned.

Hello, Starters!

The rapid development of AI takes us all by surprise, which is why we must constantly look for new ways to measure its potential... Have you heard about benchmarks?

Here’s what you’ll find today:

  • Databricks unveils its DBRX model

  • Claude 3 overtakes GPT-4's reign

  • MLCommons presents new AI benchmarks

  • OpenAI plans to pay creators on the GPT Store

  • Amazon continues to invest in Anthropic

  • And more.

Databricks has unveiled what appears to be the world's most powerful open source model, surpassing the likes of Llama 2-70B and Mistral's Mixtral on key benchmarks. DBRX harnesses 132 billion parameters, excelling in understanding, programming, and mathematics with its "mixture-of-experts" architecture.

Although DBRX has yet to surpass GPT-4, the company believes that it is more capable and cost-effective than GPT-3.5. Databricks intends that by open sourcing this groundbreaking model, they can set a new standard in the industry.

A previously unthinkable feat has been achieved by Anthropic recently; it has beaten OpenAI's GPT-4 in the Hugging Face Chatbot Arena with its latest Claude 3 Opus model.

The Hugging Face leaderboards are a valuable reference for most of those interested in AI and for researchers because they are crowdsourced, showing where the general interest lies and how the ongoing development of chatbots evolves.

For the first time, the rankings are changing, with an OpenAI model not leading the list.

MLCommons, well-known for their AI benchmarking work, has introduced a new set of tests to measure how quickly AI chips and systems can understand and respond to queries. One of the tests is aimed at language models, while the other focuses on text-to-image generation.

As we stated before, the rapid pace of AI developments demands constantly keeping track of how these models perform and which ones are achieving better results than others. This way, companies can identify blind spots and make the right improvements to their offerings.

🤝OpenAI plans to compensate "builders" in the GPT Store for their creations. The company has announced that they're testing a payment model with a small group of US-based builders, keeping their promise of rewarding them, which was initially announced with the launch of the GPT Store.

💰Amazon has announced the follow-up to its initial investment in Anthropic, adding $2.75 billion to the total. This showcases the company's commitment to staying at the forefront of the AI race by backing one of the most promising startups at the moment.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!