Deepseek

Was watching YouTube yesterday and came across a CNBC news article about a new Chinese AI called Deepseek. It was launched only this month and it was apparently developed in a few months under a budget of $20m. On some of the benchmark tests it performed better than the current US ones such as OpenAI, Gemini and Perplexity and it’s open sourced. It was also built entirely on the reinforcement learning model ie without any human involved in the training process.

The various people interviewed in the article expressed shock at this development because the thinking had been that the Chinese were about 2 years behind in the game in AI, having had a late start in AI and without having the benefit of the latest hardware due to the US ban.

So I downloaded Deepseek yesterday and explored it by asking it very general questions.

I first asked how it functioned. I then asked it what was the concept of karma, the lessons behind Hindu epic Mahabharata and what was so good about morning. I also asked if Remains of the Day was a sad story.

The wonderful thing about Deepseek is that it sets out its thought process in great detail and that is so fascinating to read and mind boggling! In fact reading the thinking process is more fun than its final answer.

Now I must see how I can use this more effectively in other areas rather than just ask general questions to test it.

PS

As I write this today on 28th Jan, US stocks took a beating after understanding the implications of what Deepseek has achieved. Nvidia down a lot.