Zuckerberg discusses AI vision: deploying multiple LLM models and launching an AI social assistant.

Recently, in computer scientist and podcast host Lex Friedman's new podcast, Mark Zuckerberg, CEO of Meta Platforms, discussed his upcoming OpenAI project.

Zuckerberg told Friedman that, compared to OpenAI's existing models, he hopes to deploy multiple large language models (LLMs) that can be fine-tuned for different life and business needs.

Zuckerberg particularly emphasized the need for human "companionship," stating that he plans to launch an OpenAI assistant to help users improve their social lives and create advertisements.

The OpenAI assistant can be both a mentor and a life coach

"Most people probably don't have as many friends as they do. You'll find tools to help people connect with each other, which is the most basic thing we want to do," Zuckerberg said. "You can imagine the OpenAI assistant reminding you of your friends' birthdays and providing suggestions on how to celebrate. In this sense, they are more like social assistants, which can help you understand what's happening in your friends' lives, how to effectively reach out to them, and how to become better friends with them."

Zuckerberg said that the roles the OpenAI assistant can take on include "mentor, life coach, cheerleader, helping you deal with all of life's challenges."

Zuckerberg believes that OpenAI is a social technology that can help people "better express themselves in situations where it is difficult to do so." According to him, language is "an imperfect way for people to express their thoughts," so OpenAI language models can help users overcome this obstacle.

Zuckerberg mentioned that his 5-year-old daughter often cries because she has difficulty expressing her thoughts, and adults can also feel frustrated because of communication difficulties.

"All of these different technologies can help us navigate the complexity of society and better express our actual feelings and thoughts, and I think that's all good."

Zuckerberg also said that the OpenAI assistant can be used for commercial purposes, and creators can now interact with brands using the OpenAI assistant.

"(Creators) see that brands want to do business with them, but they don't know how to negotiate or exit gracefully, so I think we can develop a tool that really allows them to do that. It's simple, and many people have told us they're interested in it," he said.

Zuckerberg also said that through Meta Platforms' language-image combination model, brands will no longer need to outsource ad production. In the future, if you use our service for advertising, do you need advertising creativity? No, you just need to tell us, "Yes, I am a dog walker, and I am willing to walk dogs. Please help me find a suitable employer," and then we will make the most effective advertisement. Set a goal for the system, and it will help you find the right person," he said.

Zuckerberg: No need to worry too much about tail risks

Meta Platforms officially joined the global OpenAI competition in February. At that time, it released the LLaMA language model, which, although not as large as OpenAI3, performed comparably well on many tasks. However, a week after its release, the model was suddenly leaked on 4chan, leading to a variety of open source alternatives to OpenAI.

The leak also sparked a debate about the misuse of the LLaMA model.

In the podcast, Zuckerberg said that Meta Platforms' development of OpenAI was open and academic. He said he believed that open source technology helps "release a lot of innovation and valuable feedback in the ecosystem," and he values "community-driven training."

Zuckerberg also refuted concerns about the risks associated with OpenAI.

"I've spent a lot of time thinking about OpenAI technology from this perspective. Over time, I think we need to think more because we may be getting close to a kind of superintelligence. I think everyone working on these projects today is very clear that we haven't reached that level yet."

"People are already worried that OpenAI tools will be used to do all kinds of harmful things we already know about. This will be a series of huge challenges that Meta Platforms has been working to address, whether or not there is a survival crisis.

"I'm worried that, to some extent, people may be too focused on some tail risks, and we certainly can reduce the risks associated with OpenAI, but we haven't done the work we should be doing yet."