• by Laurie Sullivan  @lauriesullivan, January 27, 2025

Meta Founder Mark Zuckerberg has been touting open-source artificial intelligence (AI) technology for years. Now advertisers will understand why.

On Monday, the Chinese artificial intelligence company DeepSeek presented a ChatGPT-like AI model called R1, which has all the familiar abilities, but can operate at a fraction of the cost of OpenAI’s, Google’s or Meta’s popular AI models.

The company said it had spent just $5.6 million training its newest AI model, compared with the hundreds of millions or billions of dollars that U.S. companies have spent on building and training AI technologies.

As U.S. companies pledge billions of dollars to AI, DeepSeek-R1 released a research paper that claims to beat the industry’s leading models on several math and reasoning benchmarks. Many of the metrics like capability, cost, openness have come in at a much lower cost to run. And it has brought down the cost to train and operate these models.

Zuckerberg on Friday said in a Facebook post the company plans to build a data center “so large that it would cover a significant part of Manhattan.” The plan is to bring a gigawatt of computing power online in 2025 and is projected to end the year with more than 1.3 million graphics processing units.

At the IAM ALM conference on Sunday, Meta Vice President of Research Joelle Pineau described some of the investments and advancements the company has made in AI, saying that combining AI with metaverse would fuel innovations for advertising.

Meta's multimodal Llama model is an open-sourced technology that now has more than 85,000 derivative models from healthcare to advertising. The idea, Pineau said, is to move from word to world models that work across many segments and through a family of AI agents.

Pineau said creators and moviemakers are working with Meta on technology that will allow them to use AI to create synthetic content that includes visuals and audio. She alluded to the ability to create complete television advertisements and possibly movies by using AI and open source platforms.

“I’m expecting models to become better at reasoning and personalization,” she said. “Now these are very generic.”