Original video in Chinese.
Key Takeaway
- AI won’t replace people, but people who use AI will replace people who don’t.
- AI empowering individuals is not an equal process; in the early stage, AI’s defining trait is: the stronger you are, the stronger it gets; the weaker you are, the weaker it gets.
- A small minority (about 5%) can use AI well, and they have an attitude of “not hyping it up or dismissing it” and an “AI perspective.”
- The two key ways to learn how to use AI are: DYOR (Do Your Own Research), meaning deeply study source knowledge; and learn Python programming so you can understand AI’s underlying logic.
- The article emphasizes that the gap in cognition is bigger than the gap in technology, and that programming ability is important in the AI era.
Full Content
There is good news, and there is bad news.
The good news is: AI won’t replace you.
The bad news is: the people who use AI will.
All the vendors are shouting: AI for ALL. AI really can help everyone. But the process of AI empowering individuals is definitely not an equal one.
Especially at this early stage, AI technology has only just started to be productized, it is still very imperfect, and it is quite hard to get started with. So at this stage, AI has an eight-character characteristic:
The stronger you are, the stronger it gets; the weaker you are, the weaker it gets.
What kind of people make AI stronger? What kind of people make AI weaker?
Take using ChatGPT as an example. Among the people I’ve seen, at least 95% are like this:
They paste in a few articles without any real thought, then ask AI to generate a new article. When they get the result and see that it’s not very satisfying, they conclude:
AI is trash. It’s all just hype from capital.
Fewer than 5% of people do it like this, and they will think through two things clearly:
First, what exactly do they want? For example, what is the core content of the article, what is the structure, what is the style?
Second, how does AI execute? For example, how will it think through our instructions, what does it need in order to do the job well, and during the process, do we need to give it feedback or guidance?
This small minority has two very valuable and important qualities.
One is attitude, which in popular terms is:
Not hyping it up or dismissing it.
They neither deify AI, thinking it can do anything, nor completely deny it, thinking that if AI can’t do 100%, then it has no value.
Because right now AI can only be a Copilot, a co-pilot. The person holding the steering wheel is still the Pilot, that is, the user. So for a Pilot without a brain, no matter how strong the Copilot is, it’s useless.
The attitude of this 5% minority is: use as much as AI can do. For the parts it does well, pay for it when you should, without hesitation. For the parts it can’t do, don’t panic either—after all, we’re not model vendors, and whether AGI can be achieved is not our concern.
The second is perspective, the AI perspective.
What is a large language model? It is a container of knowledge.
The process of training a large language model is a process of compressing knowledge. The process of using a large language model is a process of decompressing knowledge. Everything else is built by expanding outward from the large language model. For example:
- What is Fine-tune? It’s giving the large language model a tutoring class so it can learn some new knowledge.
- What is RAG? It’s giving the large language model a pile of reference books, so it can look them up when needed.
- What is an Agent? It’s giving the large language model a toolbox and an operations manual, letting it officially go to work and help us get things done.
If you’ve read KK’s What Technology Wants, you’ll get the feeling: this is absolutely a form of life or an intelligent agent different from the way we normally define things. This is also the first time we are facing a complex system other than human beings.
So if you want to understand and make good use of AI, you must switch perspectives and look at it from AI’s point of view and from the system’s point of view.
If you are in that 5% minority, or if you genuinely want to learn how to use AI, then I have two suggestions here, both of which are my own experience-based takeaways.
When I first created the Knowledge Planet newtype, I shared my own experience. In fact, I had no relevant background at all, and at first I didn’t know programming or anything like that. I started completely from zero and taught myself for half a year. What I used were these two methods.
First, DYOR, Do Your Own Research.
This phrase is very popular in the crypto world, and it means: do your own research, don’t just listen to others. It also applies to learning how to use AI.
There is a harsh truth I have to say: when it comes to the gap between China and overseas in AI, the gap in cognition is even bigger than the gap in technology.
From the media to business big shots, everyone is still learning. Especially those big names. Once you really do the research, you’ll find that they only half understand things. But why do they still dare to come out and talk, dare to teach? Two reasons:
One is influence. In the face of a technological revolution at the level of AI, everyone is starting over, no matter how capable you were before or how high your status was. In order to get a head start, of course they have to step into the spotlight proactively, while their influence from the previous era still has some warmth left.
The second is learning. In the eyes of the elite, output is also a form of learning, and an especially effective one. So it looks like they are teaching you, but in fact they are just doing their homework in front of the camera.
The most advanced and most up-to-date AI content is overseas. You can only learn it yourself; no one can teach you face to face step by step.
Once you start learning, there is one thing you need to pay attention to: try your best to find the source.
For example, you see many people discussing how the high-quality data used to train large language models is running short and will limit further improvements in model performance.
If you stop there, all you get is a so-called viewpoint that is actually not very useful. If you’re willing to ask a few more questions, such as: Why do we need massive amounts of data? What exactly does a large language model learn from this data? If there isn’t enough data, can synthetic data work? Can existing large language models generate data for the next generation of large language models to train on?
If you follow the logic all the way down and find every answer, you can really understand this issue from the source level.
Don’t get hung up on whether you need to learn in a very systematic way. You just need to make each point solid. After a while, you’ll find that these points connect into a web. And behind them is the same fundamental logic.
When you get to this point, congratulations, you’ve gotten started.
Second, learn Python.
A lot of big names are hyping this up: you don’t need to learn programming anymore; everyone can be a programmer.
I can tell you very confidently: maybe that will be true a few years from now. But right now, programming is still an irreplaceable skill.
So why learn Python programming when learning AI?
My personal experience is that learning programming is like learning English: a door to a new world opens. Only when you can understand code can you see the real AI world.
For example, a lot of media outlets and big names have started hyping Agent, talking about a bunch of definitions and whatnot, making it sound dazzling. Seriously, don’t believe a single word of it. Just honestly find a piece of AutoGen or CrewAI code and read it through once, and you’ll understand what an Agent is and how it operates.
This is the difference between insiders and outsiders. Whether you cross that threshold or not makes an enormous difference.
So Python is a must. You don’t need to write a lot of code; you just need to be able to read it—that’s not hard at all. Even a middle-aged man like me, who works in communications and marketing, can learn it. What reason do you have not to be able to learn it?
That’s my two learning suggestions based on my own experience, along with the attitude and perspective needed in the process. Actually, I wanted to make this video a month ago, but after thinking about it, I figured the number of people willing to watch would definitely be small, and I might even get attacked for it. Anyway, I still made it in the end. I hope it can inspire a small number of people. See you next time!