It has been widely accepted in the AI industry that building state-of-the-art large language models is a capital-intensive and skill-demanding activity.
This is a part of the justification for the US government’s commitment to fund the $500 billion Stargate Project that President Donald Trump announced.
What is DeepSeek?
- It is a Hangzhou-based AI development company.
- The company was founded in May 2023 by Liang Wenfeng, a Zhejiang University alumnus.
- Wenfeng is also the co-founder of High-flyer, a quantitative international hedge fund that owns DeepSeek.
- At the moment, DeepSeek operates as an independent AI research lab under the High-flyer banner.
- The total funding and valuation of DeepSeek have not been disclosed, but some of the information is available on the source.
- The company will service its models through a web interface, a mobile application, and API access.
Comparison between OpenAI vs. DeepSeek
-
Both the companies have generative AI LLMs under development; however, their methods will differ.
-
OpenAI, incorporated in 2015 and based in San Francisco, California, will aim to generalize the broad AI capabilities.
-
It has also made the models, such as GPT-4o and o1, and specializes in image generation and speech recognition with DALL-E and Whisper.
-
With regards to the AP pricing, OpenAI will charge $15 per million input tokens and $60 per million output tokens for O1, which also has a very limited open-source policy.
-
The company will also incorporate supervised and instruction-based fine-tuning for the training.
-
The estimated development cost for O1 is in the hundreds of millions of dollars.
-
DeepSeek was founded in 2023 and is headquartered in Hangzhou, China.
-
It will place a focus on efficiency and open-source models.
-
Its key models are DeepSeek-V3 and DeepSeek-R1.
-
Specialty models include DeepSeek Coder and Janus Pro for coding and vision applications, respectively.
-
It will also offer an API through DeepSeek-R1 but at significantly lower pricing of $0.55/million input and $2.19/million output.
-
Unlike OpenAI, DeepSeek has a relatively open-source policy and will also use reinforcement learning for training.
-
Moreover, the company also claims the development cost for the DeepSeek-R1 was below $6 million, thereby making it a cheaper alternative.
Innovations related to training in DeepSeek
-
DeepSeek AI approaches the training of its R1 model differently from the OpenAI methodology.
-
The process takes less time, requires fewer AI accelerators, and is also cheaper to produce.
-
DeepSeek is working towards achieving artificial general intelligence, and the progress of the company is also considered a large step forward for a more advanced AI developed in the future.
-
DeepSeek will also provide clear details on the various innovations that were also incorporated into the R1 model in their research paper.
Also Read: Invest Globally with International Stock Markets
Some of the main innovations include the following:
Reinforcement learning:
DeepSeek streamlined the entire reasoning process through large-scale reinforcements.
Reward engineering
- The researchers created a rule-based incentive system for the model that surpassed neural reward models frequently placed outperforming.
- Reward engineering refers to designing a system of the incentives that can also control the AI model during training.
Distillation
DeepSeek researchers were able to condense abilities into the models as small as 1.5 billion params by using advanced knowledge transfer models.
Emergent behavior network
The most innovative aspect is emergent behavior, which is also a phenomenon of having complex reasoning patterns that is built entirely from reinforcement learning without being programmed.
Why is DeepSeek causing concerns in the US?
- DeepSeek is raising alarms in the US.
- There are many reasons for the same.
They are as follows:
Cost disruption
-
DeepSeek states that it developed its R1 model for less than $6 million.
-
Compared to OpenAI, DeepSeek AI is also more cost-effective for the users.
Technical achievement despite restrictions
-
The export of the highest performance AI accelerator and the GPU chips from the US to China is restricted.
-
However, DeepSeek has also proven that the cutting-edge AI development can be achieved without the most sophisticated US technology.
Business models threat
-
Unlike OpenAI, which is a proprietary technology, DeepSeek AI is open source and free.
-
Such moves will jeopardize the revenue model of the US companies that will also charge monthly fees for the AI services.
Geopolitical concerns
- These factors will enable the DeepSeek, a company based in China, to contest US technological superiority in the AI.
- Tech investor Marc Andreessen referred to this as AI’s Sputnik movement, by referring to the Soviet Union mid-century space race advancement.
Conclusion
DeepSeek has completely revolutionized the industry by demonstrating a fraction of what is possible.
Frequently Asked Questions (FAQs)
Q1) How DeepSeek affects the Indian stock market?
India’s Nifty IT index plunged 3.36% on January 27, 2025 and it kept slipping the next day.
Q2) What is DeepSeek AI?
DeepSeek is a Chinese artificial intelligence (AI) company that will develop open-source large language models (LLM).
Q3) Is DeepSeek banned in India?
There is not a complete ban of DeepSeek in India.
Q4) Who is the CEO of DeepSeek AI?
Liang Wenfeng is the CEO of DeepSeek AI.
Q5) Is DeepSeek safe?
DeepSeek AI assistant, while attractive, still poses significantly security risks.
About Us
Nifty Trading Academy is our academy where we teach you about the stock market as well as technical analysis. We also provide live trading sessions and upload blogs for the same.