Llama: How DeepSeek and Meta are Contributing to AI Research & AI Product Development in the Pre-Breakdown Era

There is a number of novel things that are still being worked on by Meta and DeepSeek will be implemented into Llama. DeepSeek caused a massive sell-off in AI stocks due to fears that models will no longer need as much computing power. Zuckerberg tried to dispel concerns that the billions of dollars he’s spending on GPUs will go to waste: “I continue to think that investing very heavily in CapEx and infra is going to be a strategic advantage over time.”

Zuckerberg also made sure to praise President Donald Trump. “We now have a US administration that is proud of our leading companies, prioritizes American technology winning,” and “will defend our values and interests abroad,” he said. Moments before the earnings call started, news broke that Meta is paying Trump $25 million to settle a lawsuit he brought against the company for banning his account after the January 6th insurrection. (The vast majority of the money is going to pay for Trump’s presidential library.)

A cash-printing machine is called Meta. Net profit was a staggering $20.8 billion in the fourth quarter of the year, up 43 percent from the year before, as revenue rose 22 percent to $48.39 billion. During the earnings call, CFO Susan Li said that Meta hasn’t “seen any noticeable impact” from its content policy changes on ad spending. The number of people who used Meta’s apps daily increased by 5 percent in the fourth quarter.

DeepSeek, Stargate, and the Future of Artificial Intelligence: Where Do We Stand, Where Are We Going? What Do We Know and How Does It Come From? The Last Two Weeks on Verge

The website and mobile app of DeepSeek, as well as the ability to type into a small text box and have it reply, are a lot like the one used by the popular chat app, ChatG pt. What makes it special is how it was built. The most recent major release by the startup, a reasoning model called R1, was released on January 20th, over a month after the last model V3 had been released. It quickly became clear that DeepSeek’s models perform at the same level, or in some cases even better, as competing ones from OpenAI, Meta, and Google. Also: they’re totally free to use.

But here’s the real catch: while OpenAI’s GPT-4 reported training cost was as high as $100 million, DeepSeek’s R1 cost less than $6 million to train, at least according to the company’s claims. On Monday morning, DeepSeek became the top app in the US, and it was able to hit a hole in the stock market.

There were a lot of investors who lost a lot of money in a frenzied selloff of tech stocks. When the stock market dropped 17 percent on Monday, it was the largest decline in stock market history.

There are two competing visions for the future of Artificial Intelligence, and we have a perfect portrayal of them in DeepSeek and Stargate. One is closed and expensive, and it requires placing an ever-increasing amount of money and faith into the hands of OpenAI and its partners. The other is scrappy and open source, but with major questions around the censorship of information, data privacy practices, and whether it’s truly as low-cost as we’re being told.

With a clear indication that we has entered a new phase in the arms race of artificial intelligence, it is obvious that DeepSeek and Stargate are more than just two different paths toward superintelligence, they also represent a new front in the US-China relationship. Donald Trump has wreaked havoc on foreign relations with the threat of tariffs on foreign chips.

The news cycle is moving very quickly because of a lot of going on here. So to break it all down, I invited Verge senior AI reporter Kylie Robison on the show to discuss all the events of the past couple weeks and to figure out where the AI industry is headed next.