Announcing new customized chips, Meta gambling AI, and setting a record for infrastructure update costs

On May 19th, it was reported that over the past year, social media giant Meta has been committed to cost cutting and layoffs, but the cost of updating infrastructure has set a record, in order to maintain a leading position in the AI competition.On Thursday local time, Meta released a series of new technologies, including a new custom chip that can help train AI faster, as well as a programming tool that provides programmers with advice on developing products

On May 19th, it was reported that over the past year, social media giant Meta has been committed to cost cutting and layoffs, but the cost of updating infrastructure has set a record, in order to maintain a leading position in the AI competition.

On Thursday local time, Meta released a series of new technologies, including a new custom chip that can help train AI faster, as well as a programming tool that provides programmers with advice on developing products. In addition, the company is also renovating its data centers to make deployment easierAITechnology.

Meta CEO Mark Zuckerberg said in an email statement, "This work reflects our long-term efforts and will enable us to better utilize this technology and make more progress in various fields

To accelerate recommendation algorithms on Facebook and Instagram, Meta is developing a custom accelerator chip. In addition, the company has launched a dedicatedAIA new data center with hardware design. According to Meta, they have completed the second phase of AI supercomputer construction for training large language models similar to ChatGPT.

Last year, Meta's capital expenditure reached a historic high of $31.4 billion, an increase of more than 4.5 times compared to 2017. Zuckerberg once referred to this year as Meta's "efficiency year", and analysts estimate that Meta's investment level this year will be comparable to 2022, with many of the funds being used for improvement and expansionAIinfrastructure.

Although we have encountered some difficulties in improving efficiency, this does not mean there is a direct competition between investing in AI and improving efficiency, "said Kim Hazelwood, head of MetaAI research

Some AI updates have clearly improved Meta's internal efficiency, which has led to the company laying off thousands of employees in recent months.

CodeCompose is a new development tool based on generative AI that can automatically complete programming tasks or provide suggestions for modifying code for developers. Meta stated that there are currently 5200 internal programmers using CodeCompose, among whichAI22% of the proposed coding suggestions were adopted.

Meta is also increasingly usedAITo solve the biggest business problem. Due to Apple's adjustment of privacy policies, it has become more difficult for many advertisers to accurately target their digital advertising users. Meta plans to utilizeAITo guess the user's interest. In order to compete with the short video platform TikTok, Facebook and Instagram have also started displaying account content that users are not interested in, which requires an algorithm to guess what users may be interested in.

Angelo Zino, an analyst at market research firm CFRAResearch, said in an interview that investors will seek direct evidence to prove these improvements and justify large expenditures.

Obviously, these things will take some time and cannot be achieved overnight, "Zino said of Meta's overall increase in capital expenditure. There will definitely be a lot of reviews to ensure that we can accelerate the realization of some returns in terms of revenue

When we ask AI models, they will respond, and this process is called inference, which requires a specific type of computational process. Meta has decided to develop a new chip called MTIA (MetaTraining and Inference Accelerator) to help complete specific internal tasks and supplement the insufficient number of NVIDIA GPUs (even if they have been purchased in large quantities).

Meta hopes that the MTIA chip can enhance its ability to predict more accurate and interesting original and advertising content types for users, allowing people to spend more time using applications and click on more advertisements.

The company also launched its first internal specialized integrated circuit (ASIC) specifically designed for processing video and live streaming content. Facebook and Instagram users have shared over 2 billion short videos every day. No matter what device users use to watch videos, this new processor can help users display these videos faster and require less traffic.

Alexis Bjordin, Vice President of Hardware Engineering at Meta, said, "We have successfully optimized, balanced, and accurately adjusted the first batch of chips to fit our recommendation model

The recommendation engine used by Meta's social media app is based on current AI technology, but the future operation of generative AI will rely on Meta's AI supercomputer ResearchSuperCluster. The company will use this supercomputer to train a large number ofAIModel.

MetaLLaMAModel.

MetaTechnology.LLaMAMetaAIThe model has also been shared with researchers. The code development tool CodeCompose is built based on the publicly available technical achievements of the MetaAI research team. In addition, the new inference chip will help Meta continue to support PyTorch, which is an open source AI framework created by Meta and has been transferred to the Linux Foundation to increase its independence.

Although Meta has been committed to developing AI tools for many years, Zuckerberg still chose to set the company's future on a more vague vision of virtual reality. Scott Kessler, an analyst at investment research firm Third Bridge, said that this shift has been severely scrutinized by investors, so this deep investment in AI infrastructure can help investors rebuild confidence in Zuckerberg's overall strategy.

Kessler stated that when the industry competes to integrate AI into the business, "they don't want to lose at the starting line," he said. "Compared to six to nine months ago, more people now believe this statement." (Chen Chen)