DeepSeek Shaking Silicon Valley
What is DeepSeek?
Deepseek is a Chinese AI model that was launched in 2023. Last year, in December, DeepSeek released a reasoning model (V3) that caused lots of problems, and they released the R1 model a couple of weeks back, which shattered the industry. The R1 introduction impacted the Nvidia chipmaker market hugely.
The reason behind the revolution is that DeepSeek uses budget chips, which differ from other AI competitors. Also, DeepSeek developed DeepSeekMLA (Multi-Head Latent Attention), which dramatically reduced the memory required to run AI models by compressing how the model stores and retrieves information. OpenAI’s ChatGPT 4 cost more than 100 million dollars, but DeepSeek costs just 5.6 million USD for a final training run. This statement elaborates that most advanced chips and big datacentres don’t need to create an AI companion, and any startup can start an AI model for less cost. This made chipmakers like Nvidia’s stocks go down.
I used DeepSeek R1 for a couple of days. It gives very creative and accurate answers, the same as ChatGPT. Sometimes, I feel it is better than ChatGPT. It doesn’t provide real-time updates like Gemini, Copilot, or ChatGPT. It is updated up to October 2023. I liked the web search feature. It provides very detailed results with a deep introduction. For example, if we search for a hotel, it shows the address, facilities, and summary of user reviews. Those are well structured and tailored, making it easy to make an overall decision by looking at them. My favourite part was how it thinks when a user requests things. It shows how AI thinks and outputs the answers. It is creative and interesting to read, sometimes. On the other hand, it takes so much time to output the answer.
Is it better than ChatGPT or Gemini?
Yes, it provides better answers. It has good web search capability and it has some different patterns of outputting the answers. Also, it’s free and there are no different packages or models at a price. But it has some limitations like it cannot generate images, videos, and real-time answers. Also, the model was updated till October 2023. It’s a bit slow compared to ChatGPT or Gemini. If you use AI for creativity, writing emails, and analysis, Deepseek is the best option at no cost. But if you’re more into advanced queries like image creation, video generation, and real-time updates, ChatGPT is ideal. For normal users, Deepseek is perfect.
Security of Deepseek.
As of now deepseek faces security issues, Security researcher Gal Nagil mentioned that Deepseek exposed one of its databases to the internet which allows full control over database operations, including the ability to access internal data. However, Deepseek should provide a statement about this issue and they should focus on security and put the user privacy first.
