Li Mu's Sharing on LLM Applications and Career Development

Posted on Views: Waline: Word count in article: 763 Reading time ≈ 3 mins.
This article shares the content of Li Mu's presentation at Shanghai Jiao Tong University, including his views on the application of large language models (LLMs), the idea that "with enough data, everything can be automated", as well as his insights on working for others, pursuing a Ph.D., and starting a business, the sources of motivation, and methods of self - improvement. The author also shares personal reflections.

A few days ago, I watched a sharing given by Li Mu at Shanghai Jiao Tong University. He mainly shared his views on large language models and his personal experiences.

I really like Li Mu. He is highly skilled in technology and seems down - to - earth. The [courses](https://zh - v2.d2l.ai/) he recorded are also relatively easy to understand.

Applications of Large Language Models

The following picture shows his views on the current applications of large language models.

![Li Mu’s views on LLM applications](../images/li - mu - llm - applications.png)

Here, the definition of Simple tasks is a task that can be completed within an hour, and Complex tasks refer to tasks that require more than an hour to complete.

In his opinion, for simple tasks of white - collar workers in liberal arts, large language models can already perform very well. For complex tasks, they can be done, but the results are not very satisfactory. For simple tasks of white - collar workers in science, large language models can do them, but the effect is just so - so, and they can’t handle complex tasks yet. As for blue - collar jobs, such as waiters in restaurants, large language models are currently powerless. They can’t replace them to serve dishes.

From my personal experience of using large language models, they are a bit more capable or useful than what he said. For example, when I want to develop a backend application based on Django, I can ask the large language model to make a plan first and provide the initial implementation logic. Although I can’t directly use the output of the large language model, the general steps are correct, and sometimes the output can also give me some inspiration.

During the sharing, he mentioned that although he worked at AWS for seven and a half years and then started his own business, he couldn’t afford to use AWS because it’s too expensive… I have to say that although AWS is very useful, it’s indeed expensive. Our company has also been cutting cloud costs since last year, mainly for AWS. It is said that we spent nearly 100 million US dollars a year, which is really too much.

With Enough Data, Everything Can Be Automated

Regarding large language models, Li Mu also mentioned a point that I hadn’t realized before:

![Anything with enough data will be automated](../images/li - mu - data - automated.png)

With enough data, anything can be automated.
Thinking about it carefully, it’s actually quite scary. Essentially, our work is about generating data, such as the code we write, comments, documents, work chat records, PPTs, etc. In the future, when there is enough data in a certain field, will we be replaced?

Working for Others, Pursuing a Ph.D., and Starting a Business

Li Mu also shared his views on pursuing a Ph.D., working in a company, and starting a business.

![Li Mu’s goals and driving forces](../images/li - mu - basic - goals.png)

From working in a company, pursuing a Ph.D. to starting a business, the driving force required becomes stronger and stronger. Otherwise, it’s very difficult to persevere.

Regarding working in a company, he listed some advantages and disadvantages:

![Li Mu’s analysis of the advantages and disadvantages of working in a company](../images/li - mu - dagongren.png)

I deeply feel the last point of the disadvantages, that is, “within the company, superiors simplify the world for subordinates layer by layer. The longer you stay, the less you learn”, but I don’t completely agree.

I think that if a company or project has entered a stable stage and develops slowly, then it’s true that the longer you stay, the less you learn. But if it has been developing rapidly, there are still opportunities to learn a lot of new things. And in this case, generally, the salary will also increase faster.

Sources of Motivation

He also shared his insights on motivation.

![Li Mu’s insights on motivation](../images/li - mu - motivation.png)

I feel that in the subconscious of us Chinese people, “desire” is a derogatory term, and we generally don’t talk about this topic directly. But Li Mu is very straightforward. He shared that when he gets up early, he has a very strong desire for fame and fortune. We should face our desires directly. In many cases, desire is a good driving force.

Continuous Self - Improvement

He also shared a method of his own for continuous self - improvement. After reading it, I sigh that there are reasons why outstanding people are outstanding.

![Li Mu’s method of self - improvement](../images/li - mu - tishengziwo.png)

I didn’t have much awareness of the “review” mentioned in the picture until around 2020, and it wasn’t regular, nor did I have such a complete train of thought. I should learn from Li Mu in the future.