Meta plans to release a commercial version of its artificial intelligence model, which will enable startups and enterprises to build custom software based on the technology.
The move will allow Meta to compete with Microsoft-backed OpenAI and Google, which are making strides in the race to develop generative AI. Able to create text, images, and code, this software utilizes large scale language models (LLMs) that are trained on vast amounts of data and require enormous computational power.
Meta released its own language model, known as LLaMA, to researchers and academics earlier this year, but the new version will be more widely available and will allow companies to customize it, according to three people familiar with the plan. Stated. A release is expected soon, according to one of the people involved.
Meta says its LLM is “open source.” This means that the details of the new model will be released to the public. This contrasts with the approaches of competitors such as OpenAI. OpenAI’s latest model, GPT-4, is a so-called black box where the data and code used to build the model are not available to third parties.
“The competitive landscape for AI will completely change in the coming months, maybe weeks. In fact, there will be open source platforms that are as good as they are not,” said the company’s vice president and chief AI scientist. speaks. Meta, Yann Lecan, said at a conference in Aix-en-Provence last Saturday.
Meta’s imminent release comes amid an intensifying race among Silicon Valley tech groups to establish itself as a dominant AI participant.
Writing in the Financial Times this week, Nick Clegg, Meta’s international director, praised the open-source approach’s strengths, saying that “openness is the best antidote to the fears around AI.” . But the move will also help Meta in its efforts to catch up to its rivals, as an open model will allow companies of all sizes to improve their technology and build applications on top of it.
Meta has been researching and developing AI for more than a decade, but the November release of OpenAI’s conversational chatbot ChatGPT has prompted other big tech groups to launch similar products. seems to be in a disadvantageous position.
“The goal is to undermine OpenAI’s current dominance,” said one person familiar with Meta’s high-level strategy.
Mr. Mehta declined to comment.
Meta’s technology is open source and currently free, but the company will give enterprise customers the ability to fine-tune the model to their needs using their own data, according to two people familiar with the matter. It is said that he was considering it. A source said there are no plans to charge at this time, and that Meta will not be charging for future releases.
Joelle Pineau, vice president of AI research at Meta, declined to comment on the development of the new AI model or how it will be monetized, but said: [open source], does not completely disclaim intellectual property in the work. “
“We weren’t shy about the fact that we wanted to use these models.” [in our] product,” she added.
In 2021, CEO Mark Zuckerberg announced a shift to building an avatar-filled digital world known as the Metaverse, a project that has spent more than $10 billion annually. . This costly ambition has proven unpopular with investors, and Meta has recently been scrambling to increase its investment in AI.
Earlier this year, the social networking giant set up a generative AI division led by Chief Product Officer Chris Cox. Pinault said Cox’s team is involved not only in the research side of AI, but also in product development, “creating a whole new business.”
Zuckerberg and other executives have suggested pushing the development of multiple AI chatbots for individuals, advertisers and businesses across LLM-powered meta-platforms Instagram, WhatsApp and Facebook.
Benefits of the open-source model include higher user adoption and more data input for AI to process. The more data an LLM has, the more powerful its capabilities.
Additionally, the open source model allows researchers and developers to find and address bugs, simultaneously improving technology and security. Meanwhile, tech companies like Meta have come under years of intense scrutiny over various privacy and misinformation scandals.
Giving away software for free may seem like the opposite of making money, but experts believe companies can use this strategy to capture new markets.
“Meta has realized that they are lagging behind the current AI hype cycle. will,” said one official. company thinking.
Still, open-source AI has obvious risks and can be shaped and abused by bad actors. For example, child safety groups report an increase in AI-generated child sexual abuse images online.
The researchers also found that a previous meta-AI model, BlenderBot 2, released in 2021, was disseminating false information. Meta said that while BlenderBot 3 has been made more tolerant of this content, users have noticed that they are still generating false information.
There are also regulations and legal risks related to intellectual property and copyright. On Monday, comedian and actor Sarah Silverman filed a lawsuit against Meta and OpenAI, claiming her work was used to train models without her consent.
Meta released an open-source model, LLaMA, to researchers in February. A month later, the file was leaked even more widely through the online forum 4chan, encouraging developers to build on top of it, violating Meta’s licensing rules which state that it must not be used in commercial products. rice field.
“This model is out there in a way that we wish it weren’t,” Pinault said.
Other AI companies, such as French startup Mistral, are also considering the possibility of releasing open-source versions of their technology. OpenAI has released open-source AI models for speech and image recognition so far, but the team is considering developing an open-source LLM, provided the risk of abuse can be reduced below a minimum threshold. Stated.
“The only choice is to decide that artificial intelligence is too dangerous a technology to keep open, or to leave it in the hands of a few companies that lock it up and manage it,” said Lucan, head of AI at Meta. said Mr. “Or conversely, an open source platform seeking contributions…from around the world.”
Additional reporting by Tim Bradshaw
