Meta Strikes Back at MS and Google with Open Source Release

Mark Zuckerberg, CEO of Meta. [Photo by Yonhap News]

Meta has officially entered the artificial intelligence (AI) competition. Compared to Microsoft (MS) and Google, Meta is considered to lag behind in AI competitiveness, so it is targeting the niche market of 'open source.'

On the 24th, Meta released an open-source large-scale multilingual speech recognition (MMS) AI model that can identify more than 4,000 spoken languages and convert about 1,100 languages from speech to text. The newly released AI model expanded the number of languages capable of speech-to-text and text-to-speech conversion from around 100 to 1,107, a tenfold increase. It is known that there are over 7,000 languages currently used on Earth, and the number of languages identifiable by voice has increased to more than 4,000, about 40 times more.

Meta's AI technology is considered to be behind competitors such as MS and Google. Even during the ChatGPT craze sparked by OpenAI last year, while other competitors rushed to showcase their AI competitiveness, Meta remained relatively quiet. However, Meta has introduced a strategy to catch up with its lagging technological competitiveness by releasing open source.

By releasing AI models as open source, a large number of users can be quickly gathered. The source code, which corresponds to the blueprint of the AI model, is made available for free, allowing users to improve and redistribute it. The more users utilize the AI model, the more vast data can be collected. Through this process, the AI model can grow on its own.

OpenAI, which triggered the ChatGPT craze, also went through a similar process. It pursued a transparent and stable AI ecosystem through the open-source ecosystem, attracting many users, which became the foundation for developing ChatGPT.

In February, Mark Zuckerberg, CEO of Meta, partially released the company’s large-scale language model LLaMA, stating, “LLaMA is intended to assist researchers and will be provided in an open-source form.”

LLaMA has 6.5 billion parameters, corresponding to synapses in the human brain. This is behind Google Bard’s 530 billion and ChatGPT’s 175 billion parameters. However, unlike Google and ChatGPT, LLaMA has no restrictions on external use, gaining positive responses from many users.

Meta appears to continue its open-source release strategy. Instead of developing massive AI models with over 100 billion parameters like its competitors, it focuses on developing AI models specialized in specific areas. Last month, Meta released an open-source AI model that easily converts drawings into animations. Based on the large user base secured this way, Meta can also build its own AI platform.

Meanwhile, as Meta focuses on AI, its metaverse business seems to be sidelined. Meta even changed its company name from Facebook to Meta to focus on the metaverse business. However, in the first quarter of this year, the metaverse division, Reality Labs, posted disappointing results with revenue down 51% compared to the same period last year.

Recent job postings also clearly show a focus on AI capabilities. Most of the new hires are for AI-related technical positions. Earlier, Meta conducted large-scale restructuring last year, cutting 11,000 employees, which was 13% of its total workforce. In March this year, it announced plans for an additional 10,000 layoffs.