(Read in 3 minutes)

Because just last month, artificial intelligence in general, represented by GPT, began to come online very powerfully. The momentum is as strong as the smartphone at the time. How much momentum does it have? Give it a try with a few examples.

On March 14th, Stanford University released a large-scale language model called “Alpaca-7B”. Its performance is comparable to GPT3.5, but importantly, this model was created by fine-tuning another model and cost just under $600 to train. Around the same time, Zhipu AI, a Chinese AI company, released an AI app called “ChatGLM-6B”. This news actually says the same thing. It used to be thought that training AI was very expensive, but now it’s becoming cheaper in some ways. You can save money by spending only two months’ worth of food.

Then, on March 15th, GPT-4 came online. We talked about this the other day, but this was just the beginning.

On March 16th, Pytorch 2.0 was released. What is this for? This is an AI training tool, like an AI trainer. This 2.0 version is twice as efficient as before. Simply put, not only is the AI itself stronger, but the trainers who train it are also better than before. The speed at which humans create large-scale AI models will increase.

Similarly, we had another checking product last week. It is Microsoft 365 Copilot released by Microsoft. This is like an office with GPT-4.

Of course, Chinese companies were not idle either. For example, on March 16th, Baidu held a press conference for “One Word”.

You don’t have to memorize this news, but just one conclusion is enough. So, within the last 10 days, AI has become stronger, cheaper, and growing faster than before.

On the night of March 21st, NVIDIA’s GTC developer conference had the theme of never missing a defining moment in AI. NVIDIA CEO Huang In-hoon said many times at the conference that this is the AI iPhone moment. In other words, AI will explode in the same way that the iPhone did back then.

Here comes a very old topic, but I have a question. Is it true that if it continues to develop at this rate, AI will take over human jobs?

There are actually many opinions on the market about this issue. However, a few days ago, a company working on OpenAI, or GPT, published a paper of its own. The underlying meaning is that it’s clear we made it up. Ask us what’s going on.

And, in OpenAI’s own words, will AI replace human jobs? The answer is not necessarily. They mostly studied Americans. According to their study, 80% of Americans and 10% of their jobs are affected by GPT. 19% of Americans say 50% of their jobs could be replaced by AI.

According to OpenAI, this impact will affect nearly every industry, but to varying degrees. The higher the educational background and income, the higher the possibility of being replaced by AI. For example, lawyers, accountants, financial analysts, etc. Simply put, this round of AI specializes in competing with knowledge workers for jobs.

However, I don’t think we should worry too much about this report. There is an obvious logical flaw, at least in the wording of the report itself. For example, the word “assistant” comes up particularly frequently, such as a bartender’s assistant, a pipe fitter’s assistant, or a roofer’s assistant. And no mention is made of the bartenders or pipefitters themselves. I think there is at least a problem with this logic. I also question the rigor of this report. I think it’s more about the functional reasoning of AI, that is, looking at what AI is capable of and seeing how far it can theoretically go. It is not professional and does not focus on vocational substitutability analysis.

Thus, are there any jobs that AI cannot replace? Yes, there are. OpenAI is said to have listed 34 types of jobs that are unaffected, generally manual labor. These include agricultural machine operators, pipe fitter assistants, dishwashers, athletes, and roofer assistants. Doesn’t this make your heart hurt a little? A lot of people say that even if you study all your life, you have to work using brute force in the end.