Meta is releasing a commercial version of its open-source artificial intelligence model Llama, the company said Tuesday, offering start-ups and other businesses a powerful free alternative to the expensive proprietary models sold by OpenAI and Google. Is.
The new version of the model, called Llama 2, will be distributed by Microsoft through its Azure cloud service and run on the Windows operating system, with Meta describing Microsoft as “our preferred partner” for the release in a blog post. Referring said.
According to Meta CEO Mark Zuckerberg’s blog post and a separate Facebook post, the model, which Meta previously provided only to select academics for research purposes, can be made available via direct download and through Amazon Web Services, Hugging Face and other providers. Will also be made available through ,
“Open source fosters innovation because it enables many developers to build with new technology,” Zuckerberg wrote. “I believe that more progress would have been made if the ecosystem was more open.”
Making sophisticated models like Llama widely available and free to businesses threatens the initial dominance established in the nascent market for generic AI software by players like OpenAI, which Microsoft supports and whose models are already in use. Offers business customers through Azure. ,
The first Llama was already competitive with the models that power OpenAI’s ChatGPT and Google’s Bard chatbot, while the new Llama has been trained on 40% more data than its predecessor, with improvements in the quality of its output. Contains over 1 million annotations by humans to fix. , Zuckerberg said.
“The commercial llama could change the picture,” said Amjad Massad, chief executive officer of software developer platform Replit. He said that more than 80% of the projects out there use OpenAI’s models.
“Any incremental improvement in the open-source model is going to eat away at the closed-source model’s market share because you can run them cheaper and have fewer dependencies,” Massad said.
The announcement follows plans by Microsoft’s biggest cloud rivals, Alphabet’s Google and Amazon, to offer business customers a range of AI models to choose from.
For example, Amazon, in addition to its family of Titan models, is marketing Access to Cloud – AI from high-profile startup Anthropic. Similarly, Google has said it plans to make Cloud and other models available to its Cloud customers.
So far, Microsoft has focused on making technology from OpenAI available in Azure.
When asked why Microsoft would support an offering that could reduce the value of OpenAI, a Microsoft spokesperson said that giving developers a choice in the type of model they use is a cloud platform for AI work. As will help in increasing your status.
internal memo
For Meta, a rich open-source ecosystem of AI technology built using its models could disrupt rivals’ plans to earn revenue from their proprietary technology, the value of which would evaporate if developers could use equally powerful tools for free. Can use open-source systems.
A leaked internal Google memo titled “We have no moat, and neither does OpenAI” stunned the tech world in May after predicting a similar scenario.
Meta is also betting that it will benefit from the advances, bug fixes and products that develop from its model that could become the default for AI innovation, as it has done over the past several years with its widely adopted Done with open source AI framework. Pyotorch.
As a social media company, Zuckerberg told investors in April, Meta effectively does more than crowd-sourcing methods to reduce infrastructure costs and maximize the creation of new consumer-facing tools. There will be profits that can attract people to its advertising-supported services. Charging a fee for access to your models.
“Unlike some of the other companies in this space, we’re not selling a cloud computing service, where we try to own the various software infrastructure we’re building,” Zuckerberg said.
“For us, it would be better if the industry standardized on the basic tools we are using and so we can benefit from improvements made by others.”
However, releasing llamas into the wild also comes with risks, as it supercharges the ease with which unscrupulous actors can create products regardless of safety controls.
In April, Stanford researchers scrapped a chatbot they had previously built for $600 using a version of the llama model because it generated distasteful text.
Meta executives say they believe public releases of technologies actually reduce security risks by harnessing the wisdom of the crowd to identify problems and build resilience into the system.
The company also says it has created an “acceptable use” policy for commercial llamas that prohibits “certain use cases,” including violence, terrorism, child abuse, and other criminal activity.
Reporting by Katie Paul; Additional reporting by Jeffrey Dustin, Crystal Hu and Yuvraj Malik; Editing by Kenneth Lee, Chizu Nomiyama and David Holmes
