Meta's Llama 2 Could Make a Greater Impact than ChatGPT
How Meta's new AI model could open the floodgates for AI applications
On July 18th, Facebook owner Meta stirred the Large Language Model (LLM) market by unveiling Llama 2. Llama 2 outshines its predecessor LlaMa (released in February 2023) and is designed for integration into commercial products.
While currently available on Microsoft Azure’s AI model catalog, Meta plans on extending Llama 2 to AWS, Hugging Face, and other providers as well.
Although not as performant as OpenAI's GPT-4, Llama 2 can be run anywhere, can be fully optimized, and is free. “Free” is a big deal, as OpenAI’s API costs have been cost prohibitive for developers. By fixing this problem overnight, Meta should open the floodgates to text-based applications, just as we saw with Stable Diffusion for image-based applications.
700 Million User Limit Boxes Out Big Tech
Meta’s licensing agreement isn’t truly open source, though, as it mandates that products with >700 million monthly active users must request a license.
Why such an arbitrary number like 700 million? Here’s my guess:
Social media competitor Snap announced that they have over 750 million monthly active users earlier this year, and it makes sense why Facebook would rather not give this language model to a competitor. Since Twitter has roughly 300 million active users, though, they could still use Llama 2!
This also means that larger companies like Apple, Amazon, and Google will not have free access to Llama 2. These companies must negotiate an agreement with Meta, just as Microsoft already has.
Open Source AI Concerns & Our Negativity Bias
Critics are expressing concerns about the potential misuse of open-source AI models, fearing they could serve as tools for disinformation, synthetic biology manipulation, and spam generation.
Such concerns are nothing new, and are fairly human. They’re emblematic of humanity's usual response to novel technology, a phenomenon shaped by our innate negativity bias. It's a survival instinct, highlighting potential threats over positive outcomes.
Consider the advent of the internet. Fears of the internet included hacking, fraud, and privacy invasion.
And guess what, all of that happened! In fact, 493 million ransomware attacks happened in 2022 alone. Yet, would we say that the internet has been a net negative on the world?
The internet's history shows how fears surrounding new technologies, while often valid, don't tell the whole story. The potential for misuse exists, as does the potential for groundbreaking positive change. Meta's (mostly) open-sourcing of Llama 2, and generative AI, are guided by this optimistic view of technological progress.
Looking Forward at Potential Applications
While we wait for those groundbreaking applications, just know that we probably won’t wait for long. Several game-changing applications arose after the open source release of Stable Diffusion. Lensa, a simple photo editing tool, came out just a few months after Stable Diffusion’s release, and was number 1 in the App Store soon after.
Another app built with Stable Diffusion, Wonder, is currently the #1 graphics and design app in the App Store.
As for Llama, imagine any idea that was cost prohibitive from OpenAI is now possible, and you’ll see the endless possibilities. Consumer applications like AI-powered assistance, or mental health apps, will likely be on the App Store soon. Enterprise applications like document analysis, and market research, will likely be powered by Llama soon as well.
Looking Forward
What’s more exciting are the ideas that we haven’t considered yet, like the iPhone, Uber, or cloud storage weren’t imagined when the internet came along.
While challenges related to the training data's source and the potential for misuse remain, Llama 2's release marks a significant step in AI development. It might indeed rival the launch of ChatGPT, and their free pricing could surpass ChatGPT in terms of total utility.
Llama 2's emergence has provided a glimpse into the future of AI—a future that is open, accessible, innovative, and driven by a community working together to push the boundaries. How this will shape the AI landscape is an exciting journey that has just begun.