West Bengal, Kolkata

DeepSeek AI Deciphered

DeepSeek AI Deciphered

Artificial Intelligence is the new revolution in the tech industry that has transformed our lives in ways previously thought unimaginable. From virtual assistants to chatbots we find AI everywhere around us, intermingled with our lives. Chat GPT, Gemini, Meta AI, Alexa, etc. have become household names. One more name is now set to be added to the list, DeepSeek.

Background of DeepSeek

DeepSeek is a Hangzhou based Chinese Artificial Intelligence (AI) startup that is focused on development and research of open source AI models. The major controlling shareholder and DeepSeek founder Liang Wenfeng is also the co-founder of quantitative hedge fund High-Flyer which is focused on bringing science to financial market analysis and investing.

Founded only a year ago in 2024, DeepSeek has revolutionized the AI industry by creating an AI chatbot that rivals the dominant US-based AIs such as ChatGPT. It released its latest AI version DeepSeek V3 in December 2024 and has since then taken the internet by storm over how effective it is compared to other AI models.

The hype around DeepSeek

Historically USA based tech companies such as Open AI, Microsoft, Google and Nvidia have been considered the forerunners in the field of artificial intelligence. Chinese search engine giant Baidu’s attempt to venture into the market was deeply unimpressive and disappointing as it failed to live up to the foreign standards on several benchmarks. But now DeepSeek AI has changed the perception around AI tech advancements in China.

Some salient features that have given a lot of attention to DeepSeek around the globe are:

Cost effective development:

DeepSeek was founded just 1 year ago with a much smaller budget than contemporary AI companies. It is reported that they have developed the functional model at a fractional cost than other AI platforms.

  • Reportedly the Chinese AI app had used only 5.58 million USD equivalent of computing power for the development of its AI model. 
  • In contrast, tech giants such as ChatGPT, Meta and Microsoft have invested billions in their AI research and development projects.

Innovative methodology: 

DeepSeek has been praised for its innovative way of “thinking” that is on par and some say even above other AI models.

  • They used reinforcement learning (RL) to explore chain-of-thought (CoT) for complex problem solving. 
  • The various architectural breakthroughs such as Mixture-of-Experts (MoE) and Multi-Head Latent Attention (MLA) enabled high performance without cutting-edge hardware.

Working with technological constraints:

The USA had placed severe restrictions on the export of computer chips to China last year. This hampered Chinese AI companies’ ability to procure computational power.

  • DeepSeek founder Liang said that they had stockpiled 10,000 NVIDIA A100 GPUs before restriction but had to complement the rest with H800 and A800 chips rather than the better H100 GPUs that have been used by other US-based AI companies. 
  • Even without using advanced chipsets with higher computational power DeepSeek was able to create a competitive AI. 

Effect on global markets:

DeepSeek’s rapid rise has led to a massive impact on global markets.

  • NVIDIA stocks plunged by 17% last week. 
  • Google parent company Alphabet and Microsoft stock prices also fell 4.03% and 2.14% respectively. 
  • US President Donald Trump has dubbed it a “wake-up call” for US tech companies and has advised them to be “laser-focused on competing to win.”

Benchmark comparisons:

  • DeepSeek has met and surpassed many benchmarks when compared to other latest AI apps such as Llama 3.1 by Meta, GPT 4.0 by OpenAI and Qwen 2.5. 
  • In the coding and maths segment, they have the best benchmark ratings compared to the other AIs.

Open source philosophy:

  • DeepSeek AI has maintained an open source philosophy focusing on community driven improvements and adoption.
  • Their lower cost and open source model have allowed smaller entities to compete with tech giants.
  • It will become much easier for various fields (healthcare diagnostic, supply chain optimization, etc.) to employ AI in their daily operations with this model.

Political backdrop: 

  • DeepSeek’s rapid rise in popularity has been dubbed China's “Sputnik moment” as the Chinese AI app has shifted the attention from the West and put China on the map as a hub for global innovation.
  • This comes at a point when China and the USA are locked in a rivalry for the top spot. 
  • The Chinese government has been very critical in data monitoring and acquisition from Chinese data based companies. There is evidence for compliance to such directives as when asked “How many states are there in India?” DeepSeek’s AI chatbot refrains from replying owing to the conflict between the Chinese and Indian governments in Arunachal Pradesh and Jammu & Kashmir.

Unique benefits of DeepSeek AI over others:

  • DeepSeek charges $ 0.14 per million input tokens $ 2.19 per million output tokens. Compared to OpenAI’s $ 15 and $ 60 respectively, DeepSeek is much cheaper to operate.
  • DeepSeek has outperformed ChatGPT in the coding and maths segment, and is much more suited for such tasks.
  • The open-source nature of DeepSeek means it is more accessible for individuals and smaller companies.
  • It is free to use for the most part. Users can just login with their google accounts and it's ready to use, without any subscription or credit cards.
  • It has broken the monopoly of Western AI apps and has reshuffled the market.
  • It has paved the way for more innovative practices using less computational power making it ideal for companies that can’t afford costly equipment to maintain its AI.