The exhausted AI engineer is fleeing from Meta

Author | Yuan Silai

Editor | Su Jianxun

Image source: Generated by Unbounded AI tool

Zuckerberg may not have thought that when he focused on the metaverse, he would miss a rare opportunity in decades.

According to recent reports from foreign media, Meta lost at least one-third of its researchers last year. Lost employees are exhausted from work and have lost confidence in the company's strategy.

There is no doubt that Meta has fallen behind in this generative AI race. It tried to catch up with the forward in a messy footsteps, but found that it was powerless.

Meta once had a strong interest in AI, but like many companies, it failed to find a commercial direction. Meta turned instead to the more promising looking Metaverse, throwing the AI at the lab scientists.

Metaverse did not bring Meta the glory it imagined. Seeing that ChatGPT is popular all over the world, and Microsoft's market value has soared to 2.5 trillion US dollars, Meta once again turned the direction of the giant wheel. In February, Zuckerberg stated that he had formed a "top-level" AI team. In an internal letter sent to employees in March, Zuckerberg stated that the company's largest investment will be placed on AI.

The performance of Meta advertising business is not good, and internal layoffs are floating. In the vortex of internal and external troubles, Meta needs AI to inject fresh blood, and it cannot and dare not be absent. However, when Zuckerberg turns around and runs towards AI, where does he put the metaverse?

A miss spanning ten years

Among the giant companies in Silicon Valley, Meta did not ignore AI. In 2013, Facebook hired Yann LeCun, the "godfather" of the AI field from New York University. He began to study convolutional network models in the 1980s and was a pioneer in the field of AI.

The joining of Yang Lichun did give Meta a reputation in the academic world, and also published some heavy papers. However, Meta faced the common problems of AI companies at that time, that is, it was difficult to find a landing scenario and commercialize the technology.

After all, Yang Lichun is an academic and is not sensitive to the business world. When the big models began to heat up in 2020, Yang Lichun categorically refused to invest in this new direction, on the grounds that the big models were too "popular" and had no "scientific value".

Yang Lichun's position is actually very common in academic circles. Scholars usually put profit after research value. After OpenAI detonated the world, Yang Lichun was still not too cold. At an online meeting of executives from small media and technology companies in early 2023, Yang Lichun commented: "In terms of the underlying technology, ChatGPT is nothing particularly innovative, nor is it revolutionary. Many research labs are using the same technology, and carry out the same work.” After being canceled by the CEO of OpenAI, Yang Lichun said more directly: ChatGPT’s grasp of reality is very superficial.

For hard technology, getting out of the laboratory to large-scale application is an insurmountable threshold. No matter how shallow or unrevolutionary Yang Lichun thinks ChatGPT is academically, at least commercially, ChatGPT has achieved unrepeatable success, and has also found a foothold for AI technology after many years.

It can be said that when Meta hesitated to invest in large models, they had missed a key opportunity to seize the opportunity.

After the big model became a hot word, Meta caught up belatedly. Once again, Zuckerberg has his core business firmly in his hands. The new Generative AI team reports directly to CPO (Chief Product Officer) Chris Cox. Cox is Zuckerberg's most trusted right-hand man and one of Meta's first 15 engineers, who has been with the company for 18 years.

Meta's technical strength is not bad, and it quickly came up with its own open source model LLaMA. But for a sprawling company of 80,000 people, drastic U-turns and reorientations are bound to cause chaos. Not long after Meta has made great efforts in the metaverse, it has begun to conquer large models, and the infrastructure has not yet been prepared. Pulling out the seedlings and encouraging them to make efforts, the result is a large-scale resignation of the scientific research team. Six of the 14 authors of LLaMa-related papers said they were leaving. Compared with nimble, energetic startups, the heavy-duty Meta is losing its appeal to AI professionals.

Choose between Metaverse and AI?

In this competition, Meta is already behind at the starting line. At the AI summit held by the White House in May, Meta was not invited, which implies that Meta is no longer a company with cutting-edge technology in the field of AI.

Of course, it would be unfair to blame Yang Lichun for Meta's lagging behind. The company was already in a period of adjustment. After experiencing a sharp drop in stock prices in 2022 and major layoffs, it finally recovered slightly in 2023. Zuckerberg described 2023 as the "Year of Efficiency" -- shutting down projects and reducing staff.

The metaverse that Meta is betting on has been bleeding. According to Meta’s financial report, its Metaverse division, Reality Labs, will lose $13.7 billion in 2022, and the loss is still expanding.

In terms of revenue, due to Apple’s change in privacy policy and other reasons, Meta’s revenue will decline by 1% in 2022, and its net profit will decline by as much as 41%.

Unlike Google, Microsoft, Android, and smart cloud services, Meta's revenue structure is very single, mainly relying on social platform advertising. After Apple revised its privacy policy, Meta could not find many solutions, and its advertising business will likely continue to decline in the future.

And 2023 happens to be the year when generative AI explodes. Meta, which is reducing costs and increasing efficiency, must spend more energy to allocate limited funds in the face of AI projects that cost money and people.

Unfortunately, Meta didn't know which way to go.

Zuckerberg has no plans to abandon the Metaverse. In the first quarter of 2023, the Metaverse division lost $4 billion, a year-on-year increase of 35%. Meta also expects that Reality Labs’ losses will continue to grow in 2023.

But at the same time, Zuckerberg also stated that he will increase investment in AI, and the amount of investment will set a record in the company's history. (single largest).

Meta's AI model is an open source model, and they expect this form to bring huge benefits like Android in the future. But obviously, in the current LLaMa will not bring in much income.

And Meta's AI products are still groping. In June, Meta finally released Music Gen, its own AI music model. When ChatGPT and Midjourney are in full swing, Music Gen is not very eye-catching. Since then, Meta has released Voicebox, a speech generation system that can generate audio reply messages. Zuckerberg also announced the launch of Meta's own chatbot.

Meta has social platforms such as Instagram, Facebook, and WhatsApp. In fact, there is no shortage of AI landing scenarios.

But Meta cannot be greedy. Regardless of AI or metaverse, they are all black holes that absorb money. When the main business could not see improvement, Meta wanted to seize two bottomless pits at the same time. Unlike Microsoft and Google, which are well-funded, Meta doesn't actually have much opportunity for trial and error. Ultimately, Meta will have to decide between AI and metaverse priorities.