Meta Platforms has launched a new research tool designed for developing artificial intelligence-based chatbots and other related technologies, aiming to assert its presence in a sector that has garnered considerable attention from competitors such as Google and Microsoft.
The tool, named LLaMA, represents Meta’s latest contribution to the field of large language models. These models, according to Chief Executive Officer Mark Zuckerberg, have demonstrated considerable potential in generating text, facilitating conversations, summarizing content, and tackling complex challenges such as mathematical problems and protein structure predictions. This announcement was shared in a Facebook post on Friday.
Currently, LLaMA is not integrated into Meta’s existing products, which include platforms like Facebook and Instagram, according to a company spokesperson. However, there are plans to provide access to the technology for AI researchers.
“Meta is committed to this open model of research,” Zuckerberg noted in his communication.
Large language models operate by processing vast amounts of digital text from various sources, including news articles and social media posts. This data is employed to train the software, which can then generate content based on prompts or queries. These models are adept at numerous tasks, such as essay writing, creating tweets, engaging in chatbot conversations, and generating coding suggestions.
The increasing adoption of this technology has sparked both interest and debate, particularly as more companies develop and test products that utilize these models. Microsoft is heavily investing in OpenAI, the developer of GPT-3, which powers the popular ChatGPT chatbot. Just this month, Microsoft rolled out a test version of its Bing search engine, enhanced by OpenAI’s chat technology, though it faced scrutiny for producing occasionally inappropriate responses.
Meanwhile, Alphabet’s Google is working on a model known as LaMDA, which stands for Language Model for Dialogue Applications. The tech giant is also testing an AI-driven chat product named Bard, which has encountered its share of technical issues.
Previously, Meta introduced a large language model called OPT-175B, but LLaMA is considered a more advanced iteration. Another model released toward the end of last year, Galactica, was withdrawn shortly after launch due to findings that it frequently provided biased and inaccurate responses.
Zuckerberg has prioritized AI within the company, frequently discussing its significance in enhancing Meta’s products during earnings calls and interviews. Although LLaMA is not currently employed in Meta’s offerings, it remains a possibility for future use. Presently, Meta utilizes AI across various functionalities, including content moderation and ranking items for user feeds.
By making the LLaMA model open-source, Meta allows external developers and researchers to better understand its mechanisms, customize it for their purposes, and collaborate on related initiatives. Last year, the Big Science and Hugging Face teams released BLOOM, another open-source large language model that aimed to increase accessibility to such technology.
© 2023 Bloomberg LP
