Meta Platforms Inc. introduced a research tool for building artificial intelligence-based chatbots and other products, seeking to create a buzz for its own technology in a field lately focused on internet rivals Google and Microsoft Corp.
The tool, LLaMA, is Meta’s latest entry in the realm of large language models, which “have shown a lot of promise in generating text, having conversations, summarizing written material and more complicated tasks like solving math theorems or predicting protein structures,” Chief Executive Officer Mark Zuckerberg said in an Instagram post on Friday.
For now LLaMA isn’t in use in Meta’s products, which include social networks Facebook and Instagram, according to a spokesperson. The company plans to make the technology available to AI researchers.
“Meta is committed to this open model of research,” Zuckerberg wrote.
Large language models are massive AI systems that suck up enormous volumes of digital text — from news articles, social media posts or other internet sources — and use that written material to train software that predicts and generates content on its own when given a prompt or query. The models can be used for tasks like writing essays, composing tweets, generating chatbot conversations and suggesting computer programming code.
The technology has become popular, and controversial, in recent months as more companies have started to build them and introduce tests of products based on the models, spotlighting a new area of competition among tech giants. Microsoft is investing billions in OpenAI, the maker of GPT-3, the large language model that runs the ChatGPT chatbot. The software maker this month unveiled a test version of its Bing search engine running on OpenAI’s chat technology, which raised immediate concerns over its sometimes-inappropriate responses.
Alphabet Inc.’s Google has a model called LaMDA, orLanguage Model for Dialogue Applications. The internet search and advertising leader is testing a chat-based, AI-powered search product called Bard, which also still has some glitches.
Meta previously launched a large language model called OPT-175B, but LLaMA is a newer and more advanced system. Another model Meta released late last year, Galactica, was quickly pulled back after researchers discovered it was routinely sharing biased or inaccurate information with people who used it.
Zuckerberg has made AI a top priority inside the company, often talking about its importance to improving Meta’s products on earnings conference calls and in interviews. While LLaMA is not being used in Meta products now, it’s possible that it will be in the future. Meta for now relies on AI for all kinds of functions, including content moderation and ranking material that appears in user feeds.
Making the LLaMA model open-source allows outsiders to see more clearly how the system works, tweak it to their needs and collaborate on related projects. Last year, Big Science and Hugging Face released BLOOM, an open-source LLM that was intended to make this kind of technology more accessible.