Meta has announced the launch of Llama 2, the commercial version of its free and open-source AI framework. Freely accessible through Microsoft’s Azure cloud service and other providers, the model was previously provided only to select academics for research purposes. The model, which can produce natural-sounding text, is a competitive alternative to the expensive proprietary models offered by Google and OpenAI. It is anticipated that Llama 2 will cause significant disruption in the emerging market for generative AI applications.
To improve the quality of its outputs, Meta’s Llama 2 has been trained on 40% more data than its predecessor, with over 1 million annotations by humans. The model is competitive in the artificial intelligence market because it can compete with more advanced solutions like Google’s Bard chatbot and OpenAI’s ChatGPT. In addition, Llama 2 is now widely available for free, lowering the barrier to entry and lowering the cost for businesses to adopt AI.
The release of Llama 2 poses a serious challenge to the market leadership of Google and OpenAI’s proprietary models. Since Llama 2 will now be freely distributed by Meta, it has the potential to eat into the market share of closed-source models by giving businesses a low-cost, low-maintenance alternative. A Google memo with the title “We have no moat, and neither does OpenAI” predicted just this outcome when it was leaked.
Furthermore, Meta is betting that its model will become the de facto default for AI innovation, just as its open source AI framework PyTorch has become the de facto standard over the past few years.
Rather than charging for access to its models, Meta, as a social media company, stands to gain more from effectively crowdsourcing ways to reduce infrastructure costs and maximize the creation of new consumer-facing tools that might attract people to its ad-supported services.
In an April investor call, Zuckerberg explained that, unlike some competitors, “we’re not selling a cloud computing service where we try to keep the different software infrastructure that we’re building proprietary.” If the industry adopts common standards for the fundamental resources we use, we can all reap the benefits of the progress made by others.
By releasing Llama 2 to a wider audience, Meta hopes to lower the barrier to entry for artificial intelligence (AI) technology among small and medium-sized enterprises. This change has the potential to democratize AI technology by making it more affordable for smaller businesses that may not have the resources to invest in expensive proprietary models.
The danger of releasing Llama 2 into the wild lies in the fact that it makes it much simpler for bad actors to create products without proper safety safeguards. In April, researchers at Stanford shut down a $600 chatbot they had created with a variant of the original Llama model because it generated inappropriate content.
Executives at Meta claim that releasing technologies to the public helps reduce safety risks because it allows the masses to help find bugs and improve the system’s resilience. To prevent “certain use cases,” such as violence, terrorism, child exploitation, and other criminal activities, the company has instituted a “acceptable use” policy for commercial Llama.
For the Llama 2 launch, Meta has chosen Microsoft as its preferred partner. The model will be made available via Windows and Microsoft’s Azure cloud service. While Llama 2 will be primarily distributed via Amazon Web Services and Hugging Face, it will also be made available for direct download.
A Microsoft spokesperson explained that the company believes that offering developers more flexibility in the types of models they use will help solidify Microsoft’s position as the de facto cloud platform for AI work, which is why Microsoft would back an offering that could potentially hurt OpenAI’s value.
Microsoft’s biggest cloud competitors, Alphabet’s Google and Amazon, have also announced plans to offer multiple AI model options to their business customers. Amazon is also selling access to Claude, an artificial intelligence developed by the well-known startup Anthropic, alongside its own Titan models. Equally, Google has stated that it intends to offer Claude and other models to its cloud customers.
First reported on Reuters
Frequently Asked Questions
Q. What is Llama 2, and what does its launch mean for the AI market?
Llama 2 is the commercial version of Meta’s free and open-source AI framework. Previously accessible only to select academics, Llama 2 is now widely available through Microsoft’s Azure cloud service and other providers. This model can generate natural-sounding text and competes with expensive proprietary models offered by Google and OpenAI. Its release is expected to disrupt the emerging market for generative AI applications.
Q. What improvements have been made to Llama 2 compared to its predecessor?
To enhance the quality of its outputs, Meta’s Llama 2 has been trained on 40% more data than its previous version, with over 1 million annotations made by humans. The model now competes with more advanced solutions like Google’s Bard chatbot and OpenAI’s ChatGPT.
Q. How does Meta plan to distribute Llama 2, and why does this pose a challenge to Google and OpenAI’s proprietary models?
Meta plans to freely distribute Llama 2, making it a low-cost and accessible alternative for businesses compared to expensive closed-source models. This approach has the potential to eat into the market share of proprietary models and challenge the dominance of Google and OpenAI in the AI market.
Q. Why is Meta choosing to release Llama 2 as an open-source model?
As a social media company, Meta aims to fuel creativity by providing more programmers access to cutting-edge tools. By offering Llama 2 as open-source, Meta hopes to unlock more progress and establish itself as the de facto default for AI innovation.
Q. What are the potential benefits and risks of releasing Llama 2 to a wider audience?
The release of Llama 2 could democratize AI technology, making it more affordable for small and medium-sized enterprises that lack resources to invest in expensive proprietary models. However, there are concerns about the potential misuse of the technology by bad actors, as seen with inappropriate content generated by the original Llama model.
Q. How does Meta plan to address safety risks associated with Llama 2’s release?
Meta aims to reduce safety risks by involving the public in finding bugs and improving the system’s resilience. The company has instituted an “acceptable use” policy for commercial Llama to prevent certain use cases, such as violence, terrorism, child exploitation, and other criminal activities.
Q. Why did Meta choose Microsoft as its preferred partner for Llama 2’s launch?
Microsoft is seen as a strategic partner for Meta, and Llama 2 will be made available via Windows and Microsoft’s Azure cloud service. By offering developers more flexibility in the types of models they use, Microsoft aims to solidify its position as the de facto cloud platform for AI work.
Q. What are other major cloud competitors like Google and Amazon doing in response to Meta’s Llama 2 release?
Google and Amazon are also planning to offer multiple AI model options to their business customers. Amazon is selling access to Claude, an AI developed by startup Anthropic, alongside its own Titan models. Google intends to offer Claude and other models to its cloud customers as well.
Featured Image Credit: Unsplash