Shulou(Shulou.com)11/24 Report--

Meta CEO Mark Zuckerberg and his lieutenants hope that other companies will be free to use and profit from the new artificial intelligence software that Meta is developing. For developers and companies that are increasingly using artificial intelligence, this decision could have a significant impact.

Meta is studying how to use the next version of its open source big language model (LLM) for commercial use, a technology that could support chatbots such as ChatGPT, according to people familiar with the matter. That could spark a frenzy among artificial intelligence developers eager to find alternatives to proprietary software sold by Google and OpenAI. This will also indirectly benefit Meta's own artificial intelligence development efforts.

In February this year, Meta released LLaMA, the first version of its open source big language model, which set off a wave in the AI field to help researchers develop alternatives to proprietary software, which has become a major source of revenue for OpenAI, the creator of ChatGPT. Although Meta's model is only used for research, it still caused a sensation in the industry. Researchers use it to develop open source large language models like Vicuna, and its functionality is close to the proprietary big language model in some ways.

Meta will benefit from the release of open source artificial intelligence models. In a conference call with analysts in April, Zuckerberg said that as developers adopt and improve these models and continue to fix security vulnerabilities, Meta will be able to integrate these improvements into its artificial intelligence model that supports consumer and advertising products.

Zuckerberg, for example, has said he wants small businesses and content creators using Facebook apps to get "AI assistants" who can automatically communicate with their fans or customers. "LLaMA or the language model behind it will basically be the engine driving this trend," Zuckerberg said in an interview with podcast host Lex Friedman (Lex friedman) last week.

However, because Meta chose to release LLaMA under a research license, the open source large language model that has been spawned so far cannot be used for commercial purposes. This limits its promotion and replication in commercial applications. While there may still be many obstacles, a commercial version of LLaMA may spur the software to be used by more people. These obstacles include gaining access to dedicated servers, which are always in short supply. In addition, it is more expensive to run such servers than typical computing tasks.

In his podcast with Friedman, Zuckerberg mentioned the internal debate about how to license the second version of LLaMA. "We have applied for a research license for the first version of LLaMA, and I think the most important thing we need to consider now is, what is the right way?" he said.

OpenAI may upend everything that Meta already faces competition in the open source artificial intelligence competition. The Abu Dhabi Institute of technological Innovation in the United Arab Emirates recently released Falcon, a powerful new open source model that has begun to be favored by open source artificial intelligence developers.

There have been reports that OpenAI could further shake up the industry if it releases a new open source big language model according to internal plans. According to people familiar with the matter, OpenAI has discussed the use of large language models for commercial purposes. However, this model is unlikely to be as good as OpenAI's state-of-the-art proprietary large language model GPT-4. GPT-4 supports ChatGPT and helps the company generate hundreds of millions of dollars in revenue each year.

But while Meta is considering commercializing the next version of LLaMA, the company could face potential legal sanctions if it is found to be using copyrighted materials to train the software.

According to several former employees, Meta had been trying to resolve legal issues before the release of the first version of LLaMA. At the time, employees on the company's legal, product and privacy teams widely discussed how to license LLaMA to developers. Last month, Yann LeCun, Meta's chief artificial intelligence scientist, was asked why Meta had not released a commercial version of LLaMA so far. He replied: "because (1) I am not the only one involved in such decisions; (2) it involves legal liability." He did not give details.

A spokesman for Meta declined to comment for this article.

Google researchers are trying to avoid copyrighted materials such as Hollywood movies and use YouTube videos and other resources to train the company's next-generation proprietary big language model, it was reported on Wednesday. Previously, many news publishers worried that Google might use materials on their websites to train proprietary large language models. The model is currently providing support for Google's chat robot Bard.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.