Welp, there’s a knockout punch that no one was really expecting. 

When Meta released its (pretty much) Open Source Llama 3 model a few days ago, it didn’t take long for it to make an impact. 

Meta’s Llama 3 internal benchmarks are SUPER impressive, showing the Openish Source model is going to be a major competitor. 

Also, it’s important to note that the 8B and 70B flavors here are really the Small and Medium varieties, and the Larger 400B parameter version of Llama 3 is still going through training.

We broke that down for you here 👇

In short, with an Open Sourced model, you can download it, tweak it, add your own data to it locally, and make it your own. 

What it means:

A lot. 

We expect this to push the other closed/proprietary models from OpenAI, Google and Anthropic to make some improvements. 

And Llama is already outpunching its weight in the LLM Chatbot Arena Leaderboard.

More importantly, Meta is taking a direct shot at ChatGPT by now offering free access directly at Meta.ai. Previously, for Llama 2, you had to have a bit of tech know-how and a powerful computer to run this model locally. 

Not anymore with Meta.ai. 

So between Google’s (somewhat capable and free version of) Gemini, Microsoft Copilot, Claude 3’s Sonnet and  Meta.ai, there’s really no reason to use the free version of ChatGPT/3.5, unless they give it a major update. 

Rarely do we say something is so good that it’s a problem. 

(At least, we haven’t said that since OpenAI’s Sora.) 

Microsoft just released a research paper and examples for its VASA-1 model, which essentially turns a single image and voice sample into a scarily realistic talking avatar. 

Unlike existing tools, VASA-1 can work with photos facing various directions and incorporates factors like eye gaze direction, head distance, and emotions for enhanced control and realism.

Granted, this tech isn’t new. There’s plenty of companies that offer similar tech, such as HeyGen, Synthesia, HourOne, D-ID and others.) 

But, the quality of Microsoft’s unreleased model is on another planet. 

What it means:

Microsoft didn’t detail plans to publicly release this. But the impact for when and how Microsoft releases this into the wild will be enormous. 

Could it revolutionize many industries? Yes. 

Could it be harmful because of its deepfake capabilities? Yes. 

We’ll be keeping a watchful eye for what’s next for VASA-1. 

Bye-bye old Atlas bot. 

Hello new Atlas bot. (With more AI and electric capabilities.) 

Boston Dynamics introduced a new fully electric version of its humanoid robot, Atlas, designed for real-world applications. 

It’s equally parts creepy, yet cool. Like, how does it move like that? 

The development of the new Atlas robot showcases the continuous evolution of AI technology in the field of robotics, paving the way for enhanced capabilities and applications in various industries.

Or just more humanoid AI-powered buts who will hopefully put away our laundry? 

The AI-powered humanoid robot race is really heating up now. 

(Wait, we kinda wanna see them all actually race.) 

We’ve talked about Figure 1 and its impressive, LLM-powered capabilities. Throw in Tesla’s Optimus, a new entry from Mentee and a few dozen others, and this space is getting hot. 

What it means:

Does this mean Terminator is coming true? 

Not necessarily. 

But, the VC and Private Equity dollars are straight flowing into humanoid robot startups, especially now as they’ve started to integrate Generative AI into their systems. That was always one of the missing pieces of the puzzle over the past decade-plus. 

LLMs really change what’s possible in the humanoid/robotics space. 

It’s worth listening to this episode with Amit Noel, the Director of Robotics at NVIDIA. 

Is Apple wearing flip flips every day to work or what? 

According to a recent report from Bloomberg, Apple might be moving ahead with plans for its own internal AI model to power the next generation of its devices. 

Straightforward, right? 

Welp, just a few weeks ago, we heard that Apple might be teaming up with Google for its next AI-powered iPhone. 

And before that?

Report that Apple was spending ‘millions of dollars a day’ on its own internal LLM. 

What it means: 

We’ve said this one and we’ll say it again thousand times. Apple is never first to the party, but it’s almost always the coolest kid there. 

And right now, Apple is the only Big Tech company not fully at the GenAI party. All eyes are on the WWDC conference this June, where we’ll see Apple announce how it’ll actually incorporate AI into its software and hardware. 

Our guess? 

If they are partnering with Google, OpenAI or Baidu for anything, it’ll be short-lived. 

On-device AI is about more than just a faster and more secure experience. It’s actually about owning data. And we doubt that Apple is gonna pass up on that opportunity in the long run. 

Rumors are swirling that we’ll be seeing OpenAI release GPT-5 today, all because of recent this post from OpenAI’s Instagram story. 

What’s got the internet conspiracy theory peeps in a dizzy? The whole 22 in that photo. And the throne. 

Because it’s OpenAI CEO Sam Altman’s birthday today (the 22nd) and that OpenAI is no longer atop the LLM Chatbot Arena Leaderboard, people are assuming this kinda cryptic post is signaling something big. 

What it means: Oh, we didn’t know Sam Altman played basketball. (Check the bottom right of that photo.

Our take (along with others on the interwebs) is that this post was a nod to Caitlyn Clark, the iconic women’s basketball player who set just about ever record and was just selected with the first overall pick in the WNBA draft around the time of OpenAI’s post. 

Her jersey number? 

You guessed it…. 22. 

Maybe that didn’t turn out as expected? 

Ted Talks collaborated with OpenAI for a glimpse into the future, and unveiled a Sora AI video on what they thought the future might look like. 

We actually don’t mind it: 

The comments on their post, though, were not as kind. In short, they got dragged. 

What it means: This is one of the first major brands to release a public Sora video for promo/marketing purposes. Expect to see other big brands in the coming weeks and months to drop AI-infused Sora videos in the near future. 

(Also, we’re happy that Ted Talks clearly stated it was Sora, even though it’s fairly obvious it’s AI generated.)