Skip to main content

Why Llama 3 is changing everything in the world of AI

In the world of AI, you’ve no doubt heard about what OpenAI and Google have been up to. And now, Meta’s Llama LLM (large language model) is becoming an increasingly important player in the game, especially with its open-source nature. Meta recently made a big splash with the launch of its Llama 3 AI model, and it’s shaken up the field dramatically.

The reasons why are multiple and varied. It’s free to use, it has a wide user base, and yes, it’s open source, to name but a few. Here’s why Llama 3 is taking the AI industry by storm and may shape its future for some time to come.

Llama 3 is really good

We can debate until the cows come home about how useful AIs like ChatGPT and Llama 3 are in the real world — they’re not bad at teaching you board game rules — but the few benchmarks we have for how capable these AI are give Llama 3 a distinct advantage.

Some of the Llama 3 developers have been sharing tests comparing the latest models against some of the leading alternatives and, in many cases, Llama 3 comes out on top. These benchmarks cover mathematics, natural language processing, coding abilities, and general knowledge and understanding.

It’s here! Meet Llama 3, our latest generation of models that is setting a new standard for state-of-the art performance and efficiency for openly available LLMs.

Key highlights

• 8B and 70B parameter openly available pre-trained and fine-tuned models.
• Trained on more…

— Ahmad Al-Dahle (@Ahmad_Al_Dahle) April 18, 2024

Don’t take their word for it, though. You can test Llama 3 against itself and a range of other AI models in the AI Arena.

A massive existing user base

When OpenAI launched ChatGPT, it had some early interested parties, but it had to grow its audience organically. It did so incredibly quickly, registering tens of millions of users following the success of its GPT 3 and GPT 4 models. But Meta’s Llama 3 is already embedded in its Meta AI, which can be used across Facebook, Instagram, WhatsApp, and Messenger, as well as on the web app versions of all these services.

That’s not millions of existing users. That’s billions. All Meta has to do to get more data fon how Llama 3 is used, or how to improve it for everyday users, is to nudge these billions of users to try it out. That’s a captive audience that few other tech companies can even come close to competing with.

It’s entirely free

If Meta CEO Mark Zuckerburg’s failed Horizons VR initiative showed us anything, it’s that Meta has no problem dumping tens of billions of dollars into projects that don’t really go anywhere. But that lack of necessary monetization is an incredible strength of a company investing in such a cutting-edge field, because where other AI developers may need to get a return on their investment relatively quickly, Meta really doesn’t need to.

With its gigantic data center infrastructure and firm foundational revenue stream from its various services, Meta is able to price Llama 3 at precisely $0 for everyone. It can afford to burn through billions of dollars worth of computing power in the quest to further develop Llama 3 and future LLMs. That makes Llama 3 more readily accessible to its billions of users than any other top-tier AI out there.

It’s open source

Where price matters for users, code transparency matters for developers. Unlike the proprietary platform AIs, like ChatGPT and its various language models, Llama 3 is open source. That means that developers can look at how it runs and make changes to it. They can iterate on top of it and take the project in their own intriguing directions. That encourages further adoption, since more developers will be able to integrate the Llama 3 model into their own projects, and it encourages better code, since more people can impact its development.

That has the potential to make Llama 3 somewhat of a standard, entrenching it for the forseeable future.

It can run locally

One of the biggest drawbacks of large language model AIs like Google’s Gemini, ChatGPT, or previous iterations of Meta’s Llama is that they take an inordinate amount of computing power — not only in training, but in actual operation. That’s why you have to pay for the privilege of using some more demanding AIs, and why the response is sometimes slow when the service is busy.

But Llama 3 can run locally on consumer laptop hardware. It requires a little jumping through hoops, and you’ll need a lot of memory to make it work effectively (the laptop used in the below example is the MacBook Pro M1 Max with 64GB of RAM memory), but that’s not much of a barrier for entry.

Run Llama 3 Locally with Open-WebUI

This video is NOT sped up. The model downloaded from @ollama (llama3:instruct), running with @OpenWebUI, on a Macbook Pro M1 Max 64G.

You can run it right now on your own machine (Windows, Linux, Mac) with 1 click.

— cocktail peanut (@cocktailpeanut) April 18, 2024

This is not the final version of Llama 3, however. It uses 70 billion parameters in its responses, but there are internal versions of Llama that utilize up to 400 billion parameters. Llama 3 is also far from the first LLM to be run locally, but it is the most capable yet, and suggests that we could see this sort of AI run in smartphones, smartglasses, VR headsets, and other devices in the near future.

And it’s only the beginning

Llama 3 is amazing, but there’s a lot it can’t do just yet. It is currently entirely text-based, which means it isn’t the multimodal AI that we’ve seen with ChatGPT and similar services. It can’t process video, or audio, or images. It can’t respond to vocal commands or speak back to you. It currently only works with the English language, too.

But it will soon be able to do all those things — and likely more. Llama 3 is a warning shot across the bow of many AI developers singaling that the time of smaller startups stealing the limelight in AI development may be over. Without the kind of major financial, technological, and infrastructure backing of a company like Meta, competing directly could be very difficult indeed.

Editors' Recommendations

Jon Martindale
Jon Martindale is the Evergreen Coordinator for Computing, overseeing a team of writers addressing all the latest how to…
OpenAI and Microsoft sued by NY Times for copyright infringement
A phone with the OpenAI logo in front of a large Microsoft logo.

The New York Times has become the first major media organization to take on AI firms in the courts, accusing OpenAI and its backer, Microsoft, of infringing its copyright by using its content to train AI-powered products such as OpenAI's ChatGPT.

In a lawsuit filed in Federal District Court in Manhattan, the media giant claims that “millions” of its copyrighted articles were used to train its AI technologies, enabling it to compete with the New York Times as a content provider.

Read more
Here’s why people are claiming GPT-4 just got way better
A person sits in front of a laptop. On the laptop screen is the home page for OpenAI's ChatGPT artificial intelligence chatbot.

It appears that OpenAI is busy playing cleanup with its GPT language models after accusations that GPT-4 has been getting "lazy," "dumb," and has been experiencing errors outside of the norm for the ChatGPT chatbot circulated social media in late November.

Some are even speculating that GPT-4.5 has secretly been rolled out to some users, based on some responses from ChatGPT itself. Regardless of whether or not that's true, there's definitely been some positive internal changes over the past behind GPT-4.
More GPUs, better performance?
Posts started rolling in as early as last Thursday that noticed the improvement in GPT-4's performance. Wharton Professor Ethan Mollick, who previously commented on the sharp downturn in GPT-4 performance in November, has also noted a revitalization in the model, without seeing any proof of a switch to GPT-4.5 for himself. Consistently using a code interpreter to fix his code, he described the change as "night and day, for both speed and answer quality" after experiencing ChatGPT-4 being "unreliable and a little dull for weeks."

Read more
2023 was the year of AI. Here were the 9 moments that defined it
A person's hand holding a smartphone. The smartphone is showing the website for the ChatGPT generative AI.

ChatGPT may have launched in late 2022, but 2023 was undoubtedly the year that generative AI took hold of the public consciousness.

Not only did ChatGPT reach new highs (and lows), but a plethora of seismic changes shook the world, from incredible rival products to shocking scandals and everything in between. As the year draws to a close, we’ve taken a look back at the nine most important events in AI that took place over the last 12 months. It’s been a year like no other for AI -- here’s everything that made it memorable, starting at the beginning of 2023.
ChatGPT’s rivals rush to market

Read more