The new version of the model, called Llama 2, will be
distributed by Microsoft through its Azure cloud service and will run on the
Windows operating system, Meta said in a blog post, referring to Microsoft as
"our preferred partner" for the release.
The model, which Meta previously provided only to select
academics for research purposes, also will be made available via direct
download and through Amazon Web Services, Hugging Face and other providers,
according to the blog post and a separate Facebook post by Meta CEO Mark
Zuckerberg.
"Open source drives innovation because it enables many
more developers to build with new technology," Zuckerberg wrote. "I
believe it would unlock more progress if the ecosystem were more open."
Making a model as sophisticated as Llama widely available
and free for businesses to build atop threatens to upend the early dominance
established in the nascent market for generative AI software by players like
OpenAI, which Microsoft backs and whose models it already offers to business
customers via Azure.
The first Llama was already competitive with models that
power OpenAI's ChatGPT and Google's Bard chatbot, while the new Llama has been
trained on 40 percent more data than its predecessor, with more than 1 million
annotations by humans to fine-tune the quality of its outputs, Zuckerberg said.
"Commercial Llama could change the picture," said
Amjad Masad, chief executive at software developer platform Replit, who said
more than 80 percent of projects there use OpenAI's models.
"Any incremental improvement in open-source models is
eating into the market share of closed-source models because you can run them
cheaply and have less dependency," said Masad.
The announcement follows plans by Microsoft's largest cloud
rivals, Alphabet's Google and Amazon, to give business customers a range of AI
models from which to choose.
Amazon, for instance, is marketing access to Claude - AI
from the high-profile startup Anthropic - in addition to its own family of
Titan models. Google, likewise, has said it plans to make Claude and other
models available to its cloud customers.
Until now, Microsoft has focused on making technology
available from OpenAI in Azure.
Asked why Microsoft would support an offering that might
degrade OpenAI's value, a Microsoft spokesperson said giving developers choice
in the types of models they use would help extend its position as the go-to
cloud platform for AI work.
Internal memo
For Meta, a flourishing open-source ecosystem of AI tech
built using its models could stymie rivals' plans to earn revenue off their
proprietary technology, the value of which would evaporate if developers could
use equally powerful open-source systems for free.
A leaked internal Google memo titled "We have no moat,
and neither does OpenAI" lit up the tech world in May after it forecast
just such a scenario.
Meta is also betting that it will benefit from the
advancements, bug fixes and products that may grow out of its model becoming
the go-to default for AI innovation, as it has over the past several years with
its widely-adopted open source AI framework PyTorch.
As a social media company, Zuckerberg told investors in
April, Meta has more to gain by effectively crowd-sourcing ways to reduce
infrastructure costs and maximize creation of new consumer-facing tools that
might draw people to its ad-supported services than it does by charging for
access to its models.
"Unlike some of the other companies in the space, we're
not selling a cloud computing service where we try to keep the different
software infrastructure that we're building proprietary," Zuckerberg said.
"For us, it's way better if the industry standardizes
on the basic tools that we're using and therefore we can benefit from the
improvements that others make."
Releasing Llama into the wild also comes with risks,
however, as it supercharges the ease with which unscrupulous actors may build
products with little regard for safety controls.
In April, Stanford researchers took down a chatbot they had
built for $600 using a version of the first Llama model after it generated
unsavory text.
Meta executives say they believe public releases of
technologies actually reduce safety risks by harnessing the wisdom of the crowd
to identify problems and build resilience into the systems.
The company also says it has put in place an
"acceptable use" policy for commercial Llama that prohibits
"certain use cases," including violence, terrorism, child
exploitation and other criminal activities. © Reuters