The AI Beat: Competition intensifies

Hello dear readers! While you're listening to the new Taylor Swift (double!) album, let's also take time to review another week of blockbuster releases — those in the enterprise tech/B2B space.

First up: Llama 3. The new large language model (LLM) from Meta has been out for less than 24 hours, but it's already making big waves in the tech industry and the AI subsector in particular, rocketing up the charts of Hugging Face 's Trending models leaderboard to quickly take the no. 1 spot.

Even Meta's social networking and AI rival, Twitter /X/ xAI owner Elon Musk, couldn't help but offer some restrained praise for the new LLM, calling it "not bad."

Why is everyone so excited about the third Llama? Two big reasons:

1. It's "open source" (or close enough, with a 700 million monthly user licensing cap) so enterprises can freely and safely use it for commercial applications without paying Meta a dime. Actually, pretty much anyone can take it and use it for whatever they want, be it individual entrepreneurs and indie developers, startups to the Fortune 500, and everyone in between (so long as they comply with the licensing terms — though again, since you can download the model on a private machine or server, who would stop you from doing what you will with it?). You could retrain it or fine-tune it for your individual business needs and sectors.
2. It's very powerful, with the 8B parameter Instruct version (there are two versions out now, a 8B and 70B one, with a 400B one in training and coming later) outclassing almost all other open source models currently available, and even some closed source/proprietary models in terms of its performance and accuracy on common AI model benchmark tests.

In addition, Meta is pushing it out in the form of its new "Meta AI" stand-alone chatbot assistant and through the search bar on Facebook , Instagram , and many of its properties, along with a built-in and updated version of its AI image generator model "Imagine" (formerly Emu), giving billions of users free* access to a model near the performance of OpenAI 's GPT-4 Turbo, which recently retook the spot as the most powerful LLM in the world after a vaguely defined update.

Yet as observers pointed out on X, it appears as though AI models are nearing a plateau in performance: coincidentally (or not) topping out near the human baseline.

We'll see if OpenAI's next model, the hotly anticipated GPT-5 (or 4.5, depending on what the company decides to release next) and the Llama 3 400B parameter LLM model can push performance up more dramatically beyond the average human.

Join Us On May 8: The AI Impact Tour - San Francisco

We’re headed back to San Francisco for the next stop of the AI Impact Tour presented by DataStax! Join us as we explore cutting-edge use cases demonstrating how businesses are leveraging generative models to personalize customer experiences and optimize business processes. Request an invite to join GenAI leaders here.

Meanwhile, competition continues to heat up in the humanoid robotics sector, with both Boston Dynamics and a new startup called Mentee Robotics revealing impressive-looking prototypes designed for commercial deployments alongside humans in factories, auto manufacturing plants, warehouses, fulfillment centers, even hospitals and homes, to assist with manual labor.

Both robots will use machine learning and AI models to interface with humans, take directions, navigate the physical world and carry out tasks.

Last week, another robotics startup founded by a former Amazon exec, Cobot - Collaborative Robotics, Inc , announced a $100M raise, though it has yet to publicly show off its non-humanoid warehouse and package handling bot. What do you think it might look like?

That's all for this week! Thanks for reading and subscribing. Like, share, etc.

Read more

*As Andrew Lewis remarked 14 years ago:

Though in this case, it's unclear just how Meta would monetize customers of its free Meta AI, except through more targeted advertising -- already the business model of Insta and FB.