ChatGPT's breakthrough does not mean a comprehensive breakthrough in the field of AI

Recently, because of the capabilities brought by the product ChatGPT (and the subsequent large language model LLM), everyone has an irresistible imagination about the capabilities of AI. Especially in talk shows, a national science fiction essay essay competition has been launched. , and all the capabilities we expect AI to have are integrated into ChatGPT. Some of the output text is even close to the output level of ChatGPT.

This is harmful and will quickly disappoint the public, because what everyone said is not at all what ChatGPT currently has, or what features may be added in the future on the technical path.

Levels of artificial intelligence

The entire field of artificial intelligence is very broad and has many levels, including the field of large language models where ChatGPT is located, and the field of generating images from text where MidJourney and Stable Diffusion are located. Of course, there are also traditional speech recognition, text recognition, face recognition and other fields. In the past many years, each field has developed at a different pace.

In some fields, in the past forty or fifty years, we have been working hard to overcome difficulties, but we have been unable to make breakthroughs, or only made partial breakthroughs. For example, the field of natural language recognition where ChatGPT is located has always been a difficult problem in artificial intelligence before GPT-3 and ChatGPT took nearly ten years to achieve breakthroughs. After ten years of groping in the dark, starting in 2020, A qualitative leap. But I need to know that it is a language model, but it cannot be generalized to a general breakthrough in the field of artificial intelligence. And more systems are still struggling to wait for breakthroughs.

I heard a lot of interviews. For example, everyone said that ChatGPT can be used to draw pictures, generate videos, predict DNA sequences, and directly diagnose and prescribe. ChatGPT makes these scenarios have a very good user interface. You can do some of the above things by chatting with the robot. For example, chatting with our Awang robot can call Stable Diffusion to draw, and you can call weather, stocks, All engines, digital humans, and other models can output video, audio, and answer real-time information. Each company can be equipped with a robot to answer questions from users or employees based on the information provided by the company, but none of these capabilities are owned by ChatGPT itself. ability. (Reply to "ChatGPT" experience in the background)

ChatGPT's capabilities

ChatGPT only solves problems that large language models can solve, although these capabilities include language understanding and output, simple logic and factual knowledge support, and there are countless amazing features we are exploring, but it itself does not process images, does not process video, does not process audio, and requires countless startups to convert these media into language models and then call the functions of ChatGPT.

Bing ChatGPT means that the search results are output by ChatGPT (Bing is responsible for the search, but ChatGPT is not responsible). We also have many expert systems responsible for medical treatment, decision-making, and ERP systems responsible for checking inventory, etc. These are systems that ChatGPT can connect together, and after getting answers from them, they are handed over to ChatGPT for output. However, they are not ChatGPT itself.

ChatGPT is a good adhesive, it is the thread in the necklace that threads the pearls, but don’t generalize its capabilities and think that it is the entire necklace.

ChatGPT is related to the best entrepreneurial direction at the moment

At the same time, we cannot be fooled into thinking that there have been breakthroughs in all AI fields just because of the hundreds of times efficiency improvement brought by ChatGPT. If there are no areas of rapid development, such as AI fields other than ChatGPT or Stable Diffusion, you don’t need to enter them for the time being.

Most companies, even technology companies like Google, can only follow up on a large scale when a technology is rapidly broken through. Being able to have enough luck and persistence to create a field is too rare. .

It is not a huge change in external technical variables. It is difficult for us to answer the question "Why was there no such company before?" All answers can only point directly or indirectly to "Others are stupid, only our team is smart." And only when technology changes dramatically does the answer become clear and convincing to all of us: "Because there was no such technology before." Therefore, if you are inspired by ChatGPT and jump into the field of AI that is not a breakthrough development field such as ChatGPT/Stable Diffusion, you may easily spend your life waiting for breakthroughs in this field.

Note: Tonight, I will record a show called "ChatGPT" with my favorite podcast host, Sarah. I don’t know if you still remember it, but Sarah and I previously collaborated on a widely circulated program called "Vernacular Web3: Will the era that is coming finally be what we think it is?" ”, I am very much looking forward to chatting with Sarah about the topic of ChatGPT. You can make a reservation below. (This live broadcast is very special. Our main purpose is to record the podcast, and at the same time live broadcast "the process of recording the podcast". The final effect of the podcast is given priority, so I will not have too much interaction. If you are interested in the process of recording the podcast Students can also come in and visit).

The above is the detailed content of ChatGPT's breakthrough does not mean a comprehensive breakthrough in the field of AI. For more information, please follow other related articles on the PHP Chinese website!