Musk was revealed to have openly called for the suspension of AI research, but secretly purchased tens of thousands of GPUs to advance the AIGC project.

On April 12, according to two people familiar with the matter, although Tesla CEO Elon Musk signed an open letter, he clearly stated that Called for a six-month moratorium on advanced artificial intelligence research, but secretly promoted a new generative artificial intelligence (AIGC) project on Twitter and purchased tens of thousands of GPUs (graphics processing units) for this purpose.

According to foreign media citing people familiar with the matter, since completing the acquisition of Twitter in October last year, Musk has carried out major reforms to the social media platform in many aspects. Recently, he purchased about 10,000 graphics processors for Twitter. Given the amount of computing work required by new technologies, technology companies often use GPUs to process large artificial intelligence model data.

People familiar with the matter said Musk’s artificial intelligence project is still in its early stages, although the purchase of so much additional computing power shows that he has high expectations for the project. However, Musk has previously criticized the recent development of AIGC technology, saying that the technology is so powerful that it requires regulation to ensure that it operates in a manner consistent with the "public interest."

According to a person familiar with the matter, the artificial intelligence project promoted by Musk at Twitter involves a large language model (LLM). LLM is still essentially a kind of artificial intelligence that needs to be trained on large amounts of data in order to eventually create new, seemingly quite complex content and text on its own. There is a huge amount of data available on Twitter to train this type of AI model. For example, ChatGPT creator OpenAI previously had access to Twitter data for training purposes, although Musk said he blocked the practice in December.

Musk also found new artificial intelligence talents for Twitter. As of early March, he had officially hired engineers Igor Babuschkin and Manuel Kroiss from DeepMind, Alphabet's artificial intelligence research subsidiary. Since at least February, Musk has been in contact with people in the artificial intelligence field, including Babushkin, hoping to start his own artificial intelligence project to compete with ChatGPT.

People familiar with the matter said it is unclear what Twitter will use AIGC for. One use could be to improve search, a feature that Musk has complained about as a poor experience. He even hired security developer George Hotz to work on a three-month project to "fix" the search functionality. However, Hotez left after just one month on the job.

Informed sources also revealed that another possible use of Twitter AIGC is advertising, which is the largest pillar of social media platform revenue. Twitter has struggled to attract advertisers since Musk took over and changed many of its policies. After training, AIGC has the ability to create new advertising images and text to target specific audiences.

Currently, GPUs can be very expensive. NVIDIA, which has an estimated 95% market share, makes GPUs for large AI models priced at $10,000. People familiar with the matter said Musk has repeatedly said that Twitter's financial foundation is not stable enough, but he likely spent tens of millions of dollars to add so many new GPUs.

The new project is expected to operate out of one of Twitter's two remaining data centers, most likely the one in Atlanta, the people added. Another data center is located in Oregon. Previously, Twitter’s main data center was located in Sacramento and operated by NTT, but Musk finally suddenly decided to shut it down at the end of December last year. Twitter's head of engineering, Behnam Rezaei, also resigned.

The above is the detailed content of Musk was revealed to have openly called for the suspension of AI research, but secretly purchased tens of thousands of GPUs to advance the AIGC project.. For more information, please follow other related articles on the PHP Chinese website!