Exploring DeepSeek AI: What ChatGPT and Google Gemini Think

Frederick Nyame

January 29, 2025

A new Chinese AI assistant, DeepSeek, has gone viral. Many reports say its large language model (LLM) competes with ChatGPT 4o, Claude 3.5, and Llama 3.1 at a much lower cost. This has caused excitement on social media, especially on X, as people discuss what this means for AI development.

The Cost of AI Development

Big tech companies have long believed that building advanced AI needs massive computing power and huge investments. OpenAI, which created ChatGPT, raised $6.6 billion in October at a $157 billion valuation. The company plans to build AI agents like Operator, which can make transactions for users. Last week, OpenAI announced Stargate, a $500 billion project with SoftBank, Oracle, Nvidia, and Microsoft. This project will build massive data centers and employ 100,000 people.

But DeepSeek claims it trained its latest AI model for just $5.6 million. In comparison, OpenAI reportedly spent $3 billion on training and $4 billion on computing power last year.

DeepSeek’s Rise and Nvidia’s Fall

Investors saw DeepSeek’s low-cost AI as a game-changer. As a result, Nvidia’s stock fell 17%. Meanwhile, the DeepSeek app became the top download on the Apple App Store in the US, UK, Australia, Canada, Singapore, and China. Bloomberg reports that DeepSeek has reached 1.6 million downloads. However, it remains unclear whether it can maintain its lead. ChatGPT, with over 300 million weekly active users, is still the most widely used AI assistant.

Mixed Reactions from AI Leaders

Several AI leaders reacted to DeepSeek’s success. Aravind Srinivas, CEO of Perplexity AI, congratulated DeepSeek. He posted on X, saying, “For a while, it wasn’t clear who would beat ChatGPT first. We reached #8 last year. Looking forward to using their models this year.”

Join GizChina on Telegram

Box CEO Aaron Levie also commented. He warned against strict AI regulations. He said, “DeepSeek is a reminder that breakthroughs can happen anywhere. If AI regulation is too strict, companies will struggle to keep up.”

But some experts question DeepSeek’s claims. Reports suggest the company only used 2,000 Nvidia H800 chips, which are older models. However, Scale AI CEO Alexandr Wang disagrees. He told CNBC that Chinese labs have more powerful chips than people realize.

Read Also: DeepSeek: free ChatGPT competitor from China and how to use it

“DeepSeek probably has around 50,000 H100 chips,” Wang said. “They can’t admit it because of US export controls. While they may have more chips than expected, they will face future restrictions.”

Questions About DeepSeek’s Transparency

Some investors believe DeepSeek is not fully honest about its costs. Josh Kushner, CEO of OpenAI investor Thrive Capital, raised concerns on X. He wrote, “DeepSeek trained its model using leading AI models. They likely used restricted chips. Their terms of service say they send US customer data back to China.”

Security Risks and a Large-Scale Attack

Many people worry about AI security. Some compare DeepSeek’s rise to TikTok, which faced a US ban over data concerns. When asked if DeepSeek is safe, Google Gemini and ChatGPT both advised caution.

Gemini warned, “DeepSeek is a Chinese AI. Given today’s political climate, security and privacy risks are real. It could contain hidden malware or tools for surveillance.”

ChatGPT also advised users to be careful. It mentioned a security flaw in December that allowed hackers to take over accounts through prompt injections. DeepSeek later fixed this issue.

Security fears grew today when DeepSeek reported a “large-scale malicious attack” on its system. By the end of the day, this warning remained on its website.

DeepSeek’s rapid rise shows how fast AI technology is changing. But questions about its security, transparency, and access to powerful chips remain unanswered.

Disclaimer: We may be compensated by some of the companies whose products we talk about, but our articles and reviews are always our honest opinions. For more details, you can check out our editorial guidelines and learn about how we use affiliate links.

Source/VIA :

Forbes