Meta said on Friday that it is releasing to researchers a new large-scale language model that underlies artificial intelligence systems. A statement Published by Reuters.
Large language models categorize a wide range of texts to summarize information and create content. For example, they can ingest large amounts of information, categorize it, and derive patterns from it to answer questions with sentences that look like they were written by humans.
The move will intensify the arms race with artificial intelligence as big tech companies integrate the technology into their products to dazzle investors.
A major battle for control of the artificial intelligence technology industry began late last year with the launch of OpenAI’s Microsoft-backed ChatGPT project, prompting major tech companies; Google’s owner Alphabet to Chinese company Baidu are showcasing products related to this technology.
LAMA, which stands for Large Language Model Meta AI, will be available under a commercial license to researchers, government agencies, civil society and academia, Mita said.
The new model, which requires “significantly less” computing power than previous programs, was trained in 20 languages with an emphasis on the Latin and Cyrillic alphabet, Meta said.
Artificial intelligence has emerged as a bright spot for investments in a technology sector whose slow growth has led to widespread layoffs and reduced testing challenges.
Meta said that the “Lama” model outperforms competitors that check more variables or variables that the algorithm takes into account, adding that the version of “Lama” with 13 billion variables outperforms GPT-3″ (GPT-3).
He described a “Llama” model with 65 billion variables that competes with Google’s “Chinchilla 70B” and “PALM 540B”, which are larger than the model used by “Google” to demonstrate the “cool” robot. (cold).
In May of last year, Meta released a large language model called the OPT 175B, which was also designed for researchers and formed the basis for the new iteration of the chatbot BlenderBot.
It later introduced a model called Galactica that wrote science papers and solved math problems, but quickly pulled the demo after it produced incorrect answers.
“Award-winning beer geek. Extreme coffeeaholic. Introvert. Avid travel specialist. Hipster-friendly communicator.”
