With the rapid emergence of Chinese AI startup, DeepSeek, top AI labs across the US are getting a run for their money. According to the research paper published by DeepSeek researchers, R1 surpasses proprietary AI models such as OpenAI’s o1 reasoning model’s capabilities across math, science, and coding at a fraction of the development cost.
Meta’s AI lead, Yann LeCun, says DeepSeek AI’s immense success and broad adoption can be attributed to the model’s open-source nature. “Open-source models surpass proprietary ones,” the AI lead explained, reiterating the importance of open-source models and DeepSeek potentially bringing OpenAI’s original founding mission back to life.
However, Meta CEO Mark Zuckerberg has reportedly assembled four war rooms of engineers to investigate the secret ingredient to DeepSeek’s “overnight” success in AI (via The Information). As you may know, DeepSeek’s R1 AI is powered by DeepSeek’s open-source V3 model, which was trained with approximately $6 million.
Two teams from the four engineering war rooms will specifically focus on determining how the Chinese AI startup managed to lower the cost of developing and training its AI. Meta will reportedly leverage the information to guide the development of its next version of Llama AI. The remaining teams will be tasked with determining the data used to train DeepSeek’s AI model, potentially dictating Llama’s overhaul.
DeepSeek poses a significant challenge to US-based AI firms, including OpenAI, Anthropic, and Meta. It surpasses advanced AI models like Meta’s Llama 3.1, OpenAI’s GPT-4o, and Anthropic’s Claude Sonnet 3.5, all at a fraction of the cost.
According to two sources with close affiliations with the Facebook maker, Meta’s AI infrastructure director Mathew Oldham reportedly indicated DeepSeek’s flagship model potentially outperforms the next version of the company’s Llama AI, which is slated to ship in early 2025.
While speaking to The Information, a Meta spokesman indicated:
“We regularly evaluate all competitive models in our development process and have done so since [the company’s] Gen Al [group] was formed. Llama has been foundational in establishing the ecosystem for open-source AI models, and we couldn’t be more excited to extend this leadership with the upcoming release of Llama 4.”
Related: OpenAI invests $500 billion in Stargate to bolster AI advances
Elsewhere, Meta CEO indicated that the company is setting aside up to $65 billion to facilitate its AI advances, including constructing data centers and new hires. As you may know, Mark Zuckerberg painted a picture of a potential future where middle-level AI engineers claim software engineering jobs from professionals at Meta in 2025.