DEEPSEEK VS CHATGPT
Context: Deepseek is a new Gen AI similar to ChatGPT, with the introduction of their new model Deepseek R1 caused Nvidia to lose close to 600 billion dollars and bent OpenAI (ChatGPT's Parent Company)
Interesting Points
NVIDIA lost 600 billion dollar of market cap within few days after introduction of Deepseek R1 to put it into the perspective, market cap of Reliance Industry has market cap close to 200 billion dollars
Deepseek company was incubated in May 2023 around one and half years old only
Deepseek is open source unlike ChatGPT which is proprietary, Deepseek allow developers to experiment and customize solutions freely
What is Deepseek ?
Deepseek AI is a company based in China, which is owned by Hedge Fund known as “High-Flyer”, it recently launched Deepseek R1 model and is seen as a competitor to ChatGPT, Google Gemini and other major AI companies.
The company Deepseek was founded not so long ago but in May 2023 by Liang Wenfeng Founder of High-Flyer Hedge Fund based in China. High-flyer is a hedge fund which is involve in Quantitative Investment (It is a method used by hedge funds and investment firms to make decisions based on data, statistics, and AI models instead of human intuition to automate trading to earn profit)
What’s all the hype about?
No body expected any model such as Deepseek-R1 to be good enough to compete with ChatGPT specially from Chinese startup as China has limited access to advance chips which are required to train these models, but they did and that too in small span of time!!
AI is a power hungry and heavy capex driven Technology, for example
ChatGPT's daily power usage is nearly equal to 5,00,000 Indian households per day each using about 13.35 kilowatts per day
ChatGPT (OpenAI) has raised $12-15 billion dollars so far
but things get interesting for Deepseek ,cause the company claims to utilize only $5.6 million excluding many costs but still far lesser than other tech giants who have allocated billions of dollars.
also, Deepseek is open source unlike ChatGPT which is proprietary, Deepseek allow developers to experiment and customize solutions freely
Most importantly, they have used lesser power GPUs implying that even with less computation but with better Algo, efficient and effective AI models can be made forcing Companies’ stock which were banking on chip demand to freefall, for example major player like Nvidia Share fell by 17% loosing close to $600 billion from the market , to put it into the perspective, market cap of Reliance Industry is close to 200 billion dollars (thrice the size of market cap of reliance is lost from Nvidia Market cap within few days) after news of Deepseek.
3. Why didn’t India do it? Why can’t we build our own DeepSeek?
Protectionism!!! (government policies that restrict international trade to help domestic industries) India is an open market meaning any foreign companies can directly setup their subsidiary in India or can provide service directly and it will most likely will outperform any Indian Startup both in quality and price, but on the other hand China is a protected market meaning if a company wants to provide any service they have to enter into a Joint Venture (JV) with a Chinese Company hence demurring foreign companies to invest , hence the Chinese researchers and Chinese firms get incentivize to invest as they have both homeland develop market and no foreign threat.
4. If everything is so good, then what’s the allegation on Deepseek?
See, a model is trained (taught) through quality data, which is the real the moat of that model, now as per OpenAI allegation, Deepseek has used ChatGPT to train its Deepseek R1 model by repeated querying (It is a technique where a pretrained model known as Teacher model such as ChatGPT is asked questions repeatedly and both question and response is fed to new model known as student model like Deepseek) this process is known as Distillation now as per OpenAI terms and condition Distillation is prohibited.
Even US President Donald Trump’s advisor for Artificial Intelligence (AI) and Cryptocurrency policies David Sacks said there’s “substantial evidence” that Chinese upstart DeepSeek leaned on the output of OpenAI’s models to help develop its own technology and accused Deepseek of Distillation right after its launch and had asked trump to impose more strictive measures to restrict China of advance chips
Now interestingly the question is ChatGPT was also trained on enormous data originally so where this data came from?? and since its launch continuously it has been accused worldwide to train their model without permission even in India Adani and Reliance has joined the ongoing Lawsuit citing their respective website was scrapped to extract information to train ChatGPT
So, the question remains: who is more innocent??
Key Takeaways:
Nvidia lost $600 billion as less compute power is required to train AI models
Open-source models of Deepseek can democratize AI.
Protected markets fostered Deepseek innovation in China.
Deepsake is accused of Distillation
It is highly likely that China will face restrictions on acquiring advanced chips, in light of DeepSeek's emergence and concerns over its AI development capabilities.