© Reuters. FILE PHOTO: A woman holds a smartphone with the Meta logo in front of a new Facebook rebranding logo, Meta, in this illustrative image taken October 28, 2021. REUTERS/Dado Ruvic/Illustration/File Photo
By Yuvraj Malik and Katie Paul
(Reuters) – Meta Platforms Inc (NASDAQ:) said on Friday it was pitching to researchers a new big language model, the core software of a new artificial intelligence system, heating up an AI arms race as big companies companies are rushing to integrate technology. on their products and impress investors.
The public battle to dominate the AI technology space began late last year with the Microsoft-backed launch of OpenAI’s ChatGPT, and incited tech heavyweights from Alphabet (NASDAQ:) Inc to Baidu Inc (NASDAQ:) of China, to announce its own offerings.
Meta’s LLaMA, short for Large Language Model Meta AI, will be made available under a non-commercial license to researchers and affiliated entities in government, civil society and academia, he said in a blog post.
Large language models extract large amounts of text to summarize information and generate content. They can answer questions, for example, with sentences that can be read as if they were written by humans.
The model, which Meta says requires “much less” computing power than previous offerings, is capable of 20 languages with a focus on those with Latin and Cyrillic alphabets.
“Today’s announcement by Meta appears to be a step in testing their generative AI capabilities so they can implement them in their products in the future,” said Gil Luria, senior software analyst at DA Davidson.
“Generative AI is a new AI application that Meta has less experience with, but one that is clearly important to the future of their business.”
AI has become a bright spot for investment in the tech industry, whose sluggish growth has led to widespread layoffs and a reduction in experimental bets.
Meta said that LLaMA could outperform competitors that look at more parameters or variables than the algorithm takes into account.
Specifically, he said that a version of LLaMA with 13 billion parameters can outperform GPT-3, a recent predecessor to the model on which ChatGPT is based.
He described his 65-billion-parameter LLaMA model as “competitive” with Google’s Chinchilla70B and PaLM-540B, which are even larger than the model Google used to display its Bard chat-based search.
A Meta spokeswoman attributed the performance to more “cleaner” data and “architectural improvements” in the model that improved training stability.
In May of last year, Meta released the OPT-175B large language model, also aimed at researchers, which formed the basis of a new iteration of its BlenderBot chatbot.
He later introduced a model called Galactica, which could write scientific papers and solve mathematical problems, but quickly removed the demo after it generated bogus answers that seemed authoritative.