Since the release of DeepSeek-V3 a month ago, PASA has been closely following DeepSeek.
It wasn't until witnessing history, reaching the top of the US Apple AppStore, surpassing ChatGPT.
Looking back now, this article will introduce the usage tips of DeepSeek, as it has completely changed from the previous AI.
How to use DeepSeek?
First, let's take care of the beginners, experts can just swipe forward.
If you still don't know the entry point of DeepSeek, keep reading this section, currently you can use DeepSeek through the following ways:
Web: https://chat.deepseek.com/
App: Search for DeepSeek in major app stores and download it, remember to recognize the blue whale Logo
Usage Tips
Tip 1: Activate the Ren and Du Meridians
When using it, remember to check the button in the red box below, which is the switch to activate the Ren and Du meridians that can punch ChatGPT and kick Claude.
Tip 2: Forget the Prompt Word Templates
First, forget the long prompt word templates you have been hoarding. All prompt words are worth trying again with DeepSeek-R1, as the intelligence level of the model has been enhanced.
This is like you, as a boss, originally hiring a fresh graduate intern to work, and now it has become a colleague with 5 years of work experience. The way you assign a task and communicate with him will definitely be different.
R1 is sensitive to prompt words, using few-shot prompt words may reduce its performance, zero-shot prompt words work best.
In layman's terms, you no longer need to provide multiple examples, just clearly describe the task and steps.
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 style versions for you, just tweak it a bit and you can use it.
Tip 3: Speak Human Language
Sometimes the answers from DeepSeek might be "too forceful", very abstract and hard to understand.
Many people get stuck here and give up, thinking that this new AI is just so-so. But the correct approach is to think further about how to improve the model's output.
The answer is simple, look below.
At this time, you just need to tell it to "speak human language".
So everyone must practice hands-on, and through repeated experiences, you can figure out the model's subtleties. Actually, the model is like your boyfriend/girlfriend, you need to be attentive to figure out their subtleties.
Tip 4: Style Rewriting
Please use XX's style to rewrite the content below in Chinese, while retaining the original core content and structure.
For example, let's use Lu Xun's style to rewrite the content of the previous picture.
Or let's use Liu Run's style to rewrite the following picture of Midjourney founder's evaluation of DeepSeek.
Tip 5: Creative Fiction
I was amazed by the screenshot posted by Wei Xi on social media, the question asked was very interesting: On the day the Xuanwu Gate Incident ended, what monologue do you think Li Shimin wrote in the middle of the night?
The answer from DeepSeek-R1 was also very stunning, currently no other AI can achieve the level of DeepSeek-R1, which is exactly what the founder of Midjourney mentioned about DeepSeek's "cultural moat".
You can also send the same question to other AIs to compare.
Let's look at the answer from DeepSeek-R1:
It not only restored the emperor's tone, was familiar with historical details, but also provided hidden clues for the subsequent story. Even when I threw the copy generated by DeepSeek-R1 into an AI-generated text detection tool, the suspected AI-generated content was only 4%!
Suddenly, a phrase that Liang Wenfeng, the founder of DeepSeek, said in a previous interview flashed through my mind:
"Chinese AI cannot always follow, there must be someone who stands at the forefront of technology."
Only half a year has passed since the interview, and the whole world can no longer ignore this whale that has stepped forward first.
