Meta artificial intelligence development error, failure to use GPU in time resulted in lagging behind opponents

An internal memo shows that in the late summer of 2022, Meta CEO Mark Zuckerberg convened a team of company executives to analyze the company’s computing power for five years. hours, especially the ability to handle cutting-edge artificial intelligence.

The memo points out that despite Meta’s high-profile investments in artificial intelligence research and the company’s increasing reliance on artificial intelligence to support its growth, it has not adopted expensive artificial intelligence-optimized software and hardware systems for its main business. The social media giant's slow pace has hampered its ability to keep up with innovation as it scales. Meta will need to "fundamentally change our physical infrastructure design, software systems, and approach to delivering stable platforms" if it is to support its AI efforts.

The shakeup increased Meta's capital spending by about $4 billion per quarter, nearly double 2021, and caused it to pause or cancel plans to build data centers at four locations, the company disclosed. plan.

And Meta is facing severe financial difficulties. Since November last year, the company has been conducting unprecedented layoffs.

At the same time, the emergence of ChatGPT in November last year triggered competition among technology giants, who have released generative AI products. Five sources said that generative AI requires a lot of computing power, which intensifies the urgency of Meta expansion.

Sources revealed that Meta’s slow application of GPUs in artificial intelligence is one of the main problems. GPU chips are ideal for AI processing because they can perform a large number of tasks simultaneously, shortening the time required to process billions of pieces of data. However, GPU chips are more expensive, and chipmaker Nvidia controls 80% of the market and maintains a lead in corresponding software, sources said.

Until last year, Meta primarily used large amounts of regular CPUs to run AI workloads. The CPU is the workhorse chip in the computer world. Although it has dominated data centers for decades, it does not perform well in artificial intelligence work.

This has resulted in competitors outpacing Meta in the field of AI. They use GPU chips and have better AI software, so they can develop new AI products and services faster.

Meta has also begun using its own custom chips designed in-house to train AI, according to two sources. But in 2021, this two-pronged approach is proving to be slower and less efficient than one built with GPUs at its core. GPU chips are also more flexible than Meta's chips in running different types of models, the two sources said.

Later, as Zuckerberg pivoted the company into the Metaverse, a lack of computing power left the company unable to respond to threats, including the rise of TikTok and Apple-led ad privacy changes.

These issues caught the attention of former Meta board member Peter Thiel. In early 2022, he resigned from his position without explaining why. According to two people familiar with the matter, during a board meeting before his departure, Thiel pointed out that Zuckerberg and his executives were too focused on the development of the Metaverse and neglected Meta’s core social media business. Leaving the company vulnerable to challenges from competitors such as TikTok.

Meta had planned to launch a custom chip in 2022, but later gave up and instead ordered billions of dollars in Nvidia GPU chips that same year. At this point Meta has lagged behind peers such as Google, which in 2015 began deploying its own custom version of the GPU, called TPU.

Meta next began to reorganize the artificial intelligence department, appointing two new engineers to lead it. During that time, dozens of executives have left Meta, nearly all of which replaced its AI infrastructure leadership.

Next, Meta began revamping its data centers to accommodate the introduction of GPUs, chips that require more power and generate more heat and must be clustered closely together and between them. Dedicated network connection. The work requires significant network capacity and new liquid cooling systems to manage the cluster's heat, requiring a "complete redesign" of them.

As the work progresses, Meta begins internal plans to develop a more ambitious new chip, similar to a GPU, that can both train artificial intelligence models and perform inference. Two sources said the project will be completed around 2025.

Meta spokesman Jon Carvill declined to comment on the chip project.

Although Meta is scaling up GPUs, companies such as Microsoft and Google are promoting commercial generative artificial intelligence products, and Meta has not made much substantial progress in this regard.

Meta’s chief financial officer admitted in February that the company is not currently devoting most of its computing power to generative work. "Basically all of our artificial intelligence capabilities are used in advertising, news feeds and Reels," she said. Reels is Meta's TikTok-like short video format that is popular with young users.

Meta did not prioritize developing generative AI products until ChatGPT launched in November, according to four sources. While the company's AI research arm has been releasing technology prototypes since late 2021, it has not focused on turning them into products. However, as investor interest continues to grow, Zuckerberg in February announced the creation of a new high-level generative AI team that he said would "accelerate" the company's work in this area. .

Chief Technology Officer Andrew Bosworth also said this month that generative artificial intelligence is the area where he and Zuckerberg spend the most time, and predicted that Meta will launch new products this year.

Two people familiar with the new team said that the team’s work is in the early stages and focused on building a base model, a core program that can later be fine-tuned and adapted to different products.

Meta spokesman Carvill said the company has been developing generative artificial intelligence products on different teams for more than a year. He confirmed that the work has accelerated in the months since ChatGPT launched.

The above is the detailed content of Meta artificial intelligence development error, failure to use GPU in time resulted in lagging behind opponents. For more information, please follow other related articles on the PHP Chinese website!