A comprehensive guide to DeepSeek, a usage technique that 90% of people don’t know (recommended for bookmarking)
Since DeepSeek-V3 was released a month ago, I have been updating articles and videos related to DeepSeek because I think it is a very awesome company.
Until yesterday, history was finally witnessed, topping the US Apple App Store, surpassing ChatGPT.
Last night, it directly collapsed Nvidia and TSMC…
If you haven’t kept up with the timeline, you can go back and review this article first: DeepSeek recent news, rankings, and live demo
before coming back to this article. This article will introduce DeepSeek usage tips, because it is already completely different from previous AIs.
How do I use DeepSeek?
First, take care of the beginners, and experienced users can just scroll down.
If you don’t know the DeepSeek entry yet, continue reading this paragraph. Currently, you can use DeepSeek in the following ways:
- Website: https://chat.deepseek.com/
- App: Search for DeepSeek in the app store and download it. Remember to look for the blue whale logo
Tips
Tip 1 Remember to check the deep thinking and networked search functions
When using, be sure to check the button in the red box below. This is the switch that can punch ChatGPT and kick Claude.
Tip 2 Forget the prompt word template
Forget the long-winded prompt templates you have in your hands. All prompts are worth trying again with DeepSeek-R1, because the model’s intelligence has improved.
It’s like you, as the boss, originally hired a recent graduate intern to do the work, but now you have a classmate with 5 years of work experience doing the work. You will definitely give him different instructions for the same task.
R1 is more sensitive to prompts, and using few-shot prompts with a small sample size may reduce its performance. Zero-shot prompts with a large sample size are the most effective.
In layman’s terms, you no longer need to give it multiple examples, just describe the task and steps clearly.
For example, I want to write a New Year’s greeting for the Year of the Snake to send to my elders.
It will automatically output multiple versions in different styles for you, which you can fine-tune a little and use.
Tip 3: Speak human language
Sometimes DeepSeek‘s answers can be a bit “over the top” and the content is very abstract and difficult to understand.
Many people give up here and give up, thinking that this new AI is just like that. But the right thing to do is to think further about how to make the model’s output better.
The answer is simple, just keep reading.
At this time, you just need to tell it to “speak human language”.
So everyone must practice, and through repeated understanding, you can figure out the model’s little mind. In fact, the model is just like your boyfriend/girlfriend: you have to use your heart to figure out his/her little mind.
Technique 4: Rewriting in style
Please rewrite the following content in the style of XX in Chinese, while retaining the core content and structure of the original text.
For example, we used the style of Lu Xun to ask him to rewrite the content of the previous picture.
For example, if we use the style of Liu Run, we can ask him to rewrite the following comment from the founder of Midjourney on DeepSeek.
Technique 5: Creative fiction
I was amazed when I saw a screenshot posted by a friend on social media, and I asked a very interesting question:On the day the Xuanwu Gate Incident ended, Li Shimin wrote a soliloquy late at night. What do you think he would have written?
DeepSeek-R1’s answer is also very, very impressive. No AI can currently achieve the same level as DeepSeek-R1. This is precisely what the Midjourney founder said about DeepSeek in the comment above: “cultural moat”.
You can also send the same question to other AIs to compare.
Let’s take a look at DeepSeek-R1’s answer:
Not only does it restore the emperor’s tone, it is also particularly familiar with historical details and gives hidden clues to the follow-up story. Even if I throw the text generated by DeepSeek-R1 to an AI-generated text detection tool for detection, the suspected AI-generated content only accounts for 4%!
The data speaks for itself, of course, but your feelings are also very important.
a phrase from DeepSeek founder Liang Wenfeng that suddenly flashed through my mind:
China’s AI cannot forever follow, and someone must inevitably stand at the forefront of technology.
It’s only been half a year since the interview, and the world can no longer ignore this whale that has taken the lead.
