R1 model DeepSeek makes AI accessible to developers

R1 model is cheaper and performs well in comparison to OpenAI's o1

Pro

Image: DeepSeek

28 January 2025

Last week the US government along with Nvidia, OpenAI, Oracle and a number of financial backers announced Stargate, a $500 billion infrastructure project to turn the States into a literal AI powerhouse. Today those same players are looking at record stock market losses after a Chinese start-up revealed an AI model with comparable, arguably superior, performance to ChatGPT built for less than $6 million using mid-tier components.

At time of writing DeepSeek has outpaced ChatGPT in App Store downloads and hit Big Tech where it really hurts: its bottom line. Investors in Nvidia, the safest bet in town for at least a year, saw its share price tumble by 17% and its valuation crash by $600 billion.

The US’ vision of AI as a long haul project with plenty of heavy lifting has been bested by a smaller, lighter, cleaner operation that anyone can download and run locally for free. Some observers have joked about ChatGPT getting replaced by an AI but it’s hard to see how big money and big business can find a new business model when the competition isn’t interrested in making money.

Founded in 2023 by hedge fund manager Liang Wenfeng, DeepSeek’s goal is the creating of an artificial general intelligence capable of equalling and even surpassing human capability. While DeepSeek is nowehere near that level of capability it has earned the respect of its contemporaries.

DeepSeek’s R1 model has attracted attention for its cost-effectiveness and performance, particularly in problem solving, rivaling OpenAI’s own o1 reasoning model. The company’s training method uses a significantly smaller number of computer chips compared to leading Western AI companies, demonstrating a more efficient approach to data analysis.

Sam Altman, CEO of OpenAI, described DeepSeek’s R1 model as “impressive” and emphasised that OpenAI will continue to develop even more sophisticated models. He acknowledged the importance of AI adoption and stressed the importance of computational resources in achieving OpenAI’s goals. Altman expressed confidence in OpenAI’s ability to deliver superior models in the future and anticipated widespread surprise at the capabilities of future generations of AI.

This development has sparked discussions about the increasing accessibility and affordability of generative AI, potentially making it a widely available product. Despite its impressive performance, DeepSeek’s app was criticised due to concerns about censorship of content on politically sensitive topics such as the Tiananmen Square massacre, Uighur camps in Xinjiang province and Taiwan’s independence.

DeepSeek’s privacy policy explicitly states that user data is stored on servers in China and describes the collection of various data, including IP addresses, device IDs, text entries, audio recordings and chat history.

The question for many users is are they more comfortable handing that information over to a US entity or a Chinese one.

Business AM & TechCentral Reporters