Meta has set up four internal teams to figure out how DeepSeek, a small Chinese AI startup, managed to roll out an AI assistant that’s already being called game-changing and impressive. DeepSeek’s latest chatbot model, R1, is said to be on par with top tier AI models like ChatGPT but at a fraction of the cost. The newest large language model on the block not only optimized compute usage but also open sourced the model which makes the competition in the AI space even more intense.
Industry insiders think Meta’s Llama models might have been the inspiration for DeepSeek. Given Llama is open sourced and so widely used, it’s possible some of the design elements were borrowed. But the performance and cost efficiency of the new Chinese model has Meta surprised.
Industry insiders think Meta’s Llama models might have been the inspiration for DeepSeek. Given Llama is open sourced and so widely used, it’s possible some of the design elements were borrowed. But the performance and cost efficiency of the new Chinese model has Meta surprised.
According to insider reports, Meta’s AI infrastructure director, Mathew Oldham, has expressed concerns internally that DeepSeek’s model might even surpass the forthcoming iteration of Llama AI. This has put Meta in a race to close the gap before its own next-generation system arrives, which CEO Mark Zuckerberg previously hinted could launch in early 2025.
Inside Meta’s Response Strategy
Among the four specialized teams Meta has deployed, two are focused on deciphering how High-Flyer Capital Management - the hedge fund backing DeepSeek - managed to drastically cut training and operational costs for the model. The objective is to identify cost-reduction strategies that could be integrated into Meta’s AI projects.
A third team is examining the dataset DeepSeek used to train its model, aiming to understand whether unique data sources contributed to its efficiency. Meanwhile, the fourth group is assessing potential structural improvements for Llama based on DeepSeek’s architecture.
Despite the competitive challenge, Zuckerberg has not publicly addressed DeepSeek’s rapid emergence. However, in a recent Facebook update, he reaffirmed that Meta’s upcoming Llama iteration would set a new industry benchmark upon release. He also disclosed plans to allocate $65 billion toward AI advancements in 2025, underscoring the company’s commitment to staying ahead in the generative AI race.
Meta’s Leadership Reacts
Yann LeCun, Meta’s chief AI scientist, addressed concerns on LinkedIn, maintaining a composed stance. He argued that DeepSeek’s advancements should not be viewed as a sign of China surpassing the U.S. in AI but rather as a testament to the power of open-source models outperforming proprietary alternatives.
He emphasized that DeepSeek built upon existing open research, demonstrating the collaborative strength of open-source innovation. In his view, the ability of researchers worldwide to iterate on shared knowledge benefits the broader AI ecosystem.
With the AI landscape evolving rapidly, Meta now finds itself in an urgent race - not just to understand DeepSeek’s breakthrough but to ensure its own future models remain competitive in a space where cost efficiency and open-source strategies are increasingly shaping industry leadership.
Image: DIW-Aigen
Read next:
• Is YouTube’s Ad Moderation Failing to Protect Children from Inappropriate Content?
• Which Cryptocurrencies Are Americans Most Likely to Invest in for 2025?
• Mobile App Usage Surpasses 4.2 Trillion Hours in 2024, Revenue Growth Continues
Image: DIW-Aigen
Read next:
• Is YouTube’s Ad Moderation Failing to Protect Children from Inappropriate Content?
• Which Cryptocurrencies Are Americans Most Likely to Invest in for 2025?
• Mobile App Usage Surpasses 4.2 Trillion Hours in 2024, Revenue Growth Continues