When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

Meta creates four 'war rooms' to unravel how DeepSeek is outperforming rivals at lower costs

Meta
image via Depositphotos

DeepSeek AI has disrupted the AI landscape in the US. In just a few weeks after the launch of its AI model, DeepSeek overtook ChatGPT to become the number one free app on the App Store. Not only this, DeepSeek's rise in popularity sent shockwaves to the tech industry, leading to a $400 billion in market cap loss for NVIDIA in the US. Recently, DeepSeek launched its Janus-Pro 7B, a groundbreaking image generation model that started making headlines, as it outperformed the likes of OpenAI's DALL-E, Stability AI's Stable Diffusion, and other image generation models in several benchmarks.

The popularity of DeepSeek has caught the attention of Meta, and to understand the success of this Chinese AI startup, Mark Zuckerberg's Meta has reportedly assembled four specialed teams, referred to as "war rooms," consisting of engineers to understand how a Chinese AI startup backed by High-Flyer Capital Management has managed to achieve performance on par with or exceeding that of top competitors like ChatGPT at a fraction of the cost.

Notably, DeepSeek gained popularity after it launched the R1 model, an AI chatbot that beat ChatGPT. The company claims that it invested less than $6 million to train its model, as compared to over $100 million invested by OpenAI to train ChatGPT. Meta's war rooms will be brainstorming to find ways how to address the potential threat posed by DeepSeek's breakthrough.

Two of the four war rooms will be dedicated to understanding how DeepSeek managed to cut costs in developing and running R1 models, with hopes of applying the same strategy to Meta's own AI model, Llama. Another team will be investigating the training data that DeepSeek used. The last team will be focussing on exploring ways to redesign Llama's architecture to compete with Chinese AI technology.

Although Meta did not comment on this development, a Meta spokesperson said in a statement to The Information that:

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.

Meta is on high alert because Meta AI infrastructure director Mathew Oldham has told colleagues that DeepSeek’s newest model could outperform even the upcoming Llama AI, expected to launch in early 2025. Even OpenAI's CEO Sam Altman has responded to DeepSeek's rise and called it impressive. NVIDIA, which is one of the biggest sufferers of the sudden popularity of DeepSeek, also commended the Chinese AI and also highlighted how NVIDIA GPUs were used for DeepSeek's software.

Source: The Information

Report a problem with article
Klipsch R-51PM
Next Article

The popular Klipsch R-51PM home and PC active speaker falls back to lowest price

DeepSeek
Previous Article

Hugging Face wants to make DeepSeek R1 fully open by filling closed source gaps

Join the conversation!

Login or Sign Up to read and post a comment.

19 Comments - Add comment