DeepSeek: What lies under the bonnet of the new AI chatbot? DeepSeek founders

 

Tumbling Stock Market and the Rise of DeepSeek AI

The release of China’s DeepSeek AI-powered chatbot has sent shockwaves through the technology industry. Rapidly surpassing OpenAI’s ChatGPT as the most downloaded free iOS app in the U.S., its debut also triggered a record-breaking $600 billion (£483 billion) loss in Nvidia’s market value in just one day.

The Innovation Behind DeepSeek

The turmoil surrounding DeepSeek stems from its cutting-edge "large language model" (LLM), which boasts reasoning capabilities comparable to top U.S. models like OpenAI’s GPT-4 but at a fraction of the training and operational cost. DeepSeek has achieved this efficiency through advanced computational strategies that minimise the time and memory required to train its model, R1. According to reports, R1’s base model V3 required 2.788 million GPU hours to train at an estimated cost of under $6 million (£4.8 million), compared to the over $100 million (£80 million) investment needed for GPT-4.

Impact on Nvidia and the AI Industry

Despite the financial blow to Nvidia, DeepSeek’s training process still relied on approximately 2,000 Nvidia H800 GPUs. These chips, modified to comply with U.S. export regulations, were likely stockpiled before the Biden administration imposed tighter restrictions in October 2023. Working within these limitations, DeepSeek has devised innovative methods to maximise its hardware’s efficiency.

Reducing AI’s computational cost could also alleviate environmental concerns. Data centres powering AI models consume vast amounts of electricity and water, primarily for cooling. While AI companies rarely disclose their carbon footprint, estimates suggest that ChatGPT alone emits over 260 metric tonnes of CO2 per month—comparable to 260 flights from London to New York. If DeepSeek’s efficiency claims hold true, its advancements could set a precedent for more sustainable AI development.

A Rapid Rise to Prominence

Founded by Liang Wenfeng in 2023, DeepSeek’s meteoric rise has surprised many. The company’s success is partly attributed to its use of a "mixture of experts" model, where smaller, specialised models handle distinct tasks. This technique was also employed in Mistral AI’s Mixtral 8x7B model in 2023. Additionally, DeepSeek has openly shared some of its unsuccessful attempts to enhance reasoning, such as Monte Carlo Tree Search, providing valuable insights for future AI advancements.

Openness and the Future of AI

Unlike OpenAI’s proprietary models, DeepSeek has released its model’s "weights"—the numerical parameters derived from training—along with technical documentation. This transparency allows researchers worldwide to analyse and adapt the model, fostering a more open AI development ecosystem. However, certain critical details, such as training datasets and code, remain undisclosed.

DeepSeek’s breakthrough demonstrates that cutting-edge AI doesn’t necessarily require vast financial or computational resources. As AI development becomes more efficient, smaller companies may challenge Big Tech’s dominance in the field. Former U.S. President Donald Trump has called DeepSeek’s emergence "a wake-up call" for the U.S. tech industry. Yet, this shift may ultimately benefit Nvidia and other tech giants by increasing demand for AI-powered solutions and the hardware that supports them.

The AI landscape is evolving rapidly, with companies like DeepSeek playing an increasingly significant role. As innovation continues, the impact of smaller players on the industry should not be underestimated.

DeepSeek founders

Liang Wenfeng: The Visionary Behind DeepSeek’s AI Revolution

Liang Wenfeng, the founder and CEO of DeepSeek, is a key figure in China’s rapidly evolving artificial intelligence (AI) sector. Born in 1985 in Guangdong, China, he pursued a degree in electronics at Zhejiang University, where he cultivated a deep interest in machine learning and its applications in finance.

Before venturing into AI, Liang made his mark in the financial industry as a hedge fund entrepreneur. His transition to AI reflects a strong belief in technology’s potential to drive innovation and transform industries. Through DeepSeek, he aims to establish China as a global leader in AI, contributing to the country’s growing influence in cutting-edge technological advancements.

Let me know if you'd like any further refinements!