In the past year, the development of AI large models has been unstoppable, and it has also promoted the iteration of the entire AI field. So, what role will open source play in it?
On June 15, 2023, at the "18th Open Source China Open Source World Summit Forum", sponsored by the China Open Source Software Promotion Alliance, jointly organized by CCID Media and "Software and Integrated Circuits" magazine, the deputy director of the COPU Technical Committee Liu Ming read the speeches of Sam Altman and Elon Musk excerpted from the COPU data group, and conducted an in-depth discussion based on the emergence and epiphany phenomena of artificial intelligence under the theme of "Can large-scale artificial intelligence surpass human beings".
The following is the transcript of the speech:
Dear Mr. Lu, Chairman Sun, Jim, and friends old and new, hello!
First of all, I would like to thank Mr. Lu for assigning me to do a brief interpretation. Before reading it, I would like to share two passages:
In the first paragraph, I want to express my mood today with an ancient poem: "Meeting each other is like seeing each other for the first time". The 3-year epidemic has made us old friends finally able to meet together like a few years ago, which makes me very excited. This ancient poem expresses the philosophy behind my interpretation that things often go beyond our wishes and expectations.
In the second paragraph, I take today's opportunity to talk about the 18 years that are closely related to me. Turning 18 is a rite of passage for a person, and it means starting again for each of us. In 2004, I happened to host a session related to open source at this conference, which was also an experience that impressed me. We share the spirit and culture of open source, extending from the campus to the wider field. My best wishes and gratitude for the development of COPU (Coalition for the Advancement of Open Source).
Next, I will interpret a topic, that is, "Can artificial intelligence of large models surpass humans?". This topic is very far-reaching. It involves not only the technical level, but also the ideological level. It is also an area that we need to continue to explore. It may never come to a definitive conclusion because it covers multiple levels of technology and thinking.
In November last year, OpenAI released the ChatGPT large language model, which aroused heated discussions in global public opinion. In my impression, the upsurge on GPT happened earlier in April 2022. At that time, Mr. Lu personally assigned me to make a report on the large model. Although there was no GPT-3.5 at the time, only GPT-3, OpenAI has released a version of GPT with human participation. In that report, I also interpreted the source of GPT, and used Pengcheng Cloud Brain to conduct pure Chinese language model research.
Later, OpenAI first launched ChatGPT about the reasoning ability of machines, and GPT-3.5 also exploded. With the development of storage technology, the large model has reached a certain point, and more and more experts have observed the changes in the stage of AI from perception to cognition.
Reasoning ability is the basis for machines to generate human natural language, and the sudden emergence of reasoning ability creates conditions for the realization of human-machine dialogue in natural language. During the epidemic, it was generally believed that face recognition is a kind of perception ability, and now cognitive ability has also been developed. Sam Altman once said that for most of the GPT series, we used it as a database. However, as ChatGPT enters the primary stage of AGI (artificial general intelligence), a puzzling reasoning ability emerges. As Mr. Musk said, ChatGPT suddenly has a reasoning ability that we cannot explain. This wave has sparked much controversy, with the most authoritative voices coming from Mr. Altman and Mr. Musk. At this node, AGI (General Artificial Intelligence), including ChatGPT, suddenly surpassed humans, which formed a consensus, although some people opposed this view.
Some domestic experts, including academicians, also proposed the phenomenon of emergence and epiphany in artificial intelligence. Mr. Lu convened a seminar to discuss why there is epiphany and emergence, as well as phenomena that humans cannot understand. Due to time constraints and considering the importance of information, the COPU data group under the leadership of Mr. Lu regularly organizes data in the field of open source and artificial intelligence around the world and shares them with you. This is actually a collision of ideas and knowledge. Some people think the profile is not important, but we think it is very important. I also regularly receive materials sent to me by Mr. Lu, and we will discuss these materials regularly.
Last year, I compiled a document covering the development process of large models from GPT-3 to nearly trillion parameters. In the process, we do encounter phenomena that are difficult for humans to understand. This is worthy of our exploration. Perhaps the best way is to discuss the phenomenon of future exploration through the open source model, where everyone cooperates, like the crowd intelligence model mentioned by Academician Wang Huaimin. This is a topic that is always worth discussing.
Finally, I would like to summarize the points that Mr. Lu particularly emphasized. When the pre-trained semantic parameters in the large model are increased to a considerable extent, ChatGPT suddenly exhibits reasoning ability, capable of generating natural language and even code. General artificial intelligence is AGI, including ChatGPT, which may surpass humans. History has proved that this kind of transcendence is a spiral gene, and human intelligence is constantly intertwined and developed. I believe that, as Mr. Lu predicted, although machines temp orally surpass humans , in the future humans will surpass the intelligence of machines even more and achieve continuous surpassing.Mr. Altman and Mr. Musk humorously said that this is a puzzling phenomenon, but they understand the meaning of it in their hearts. The important thing is to explore and work together through practice in an open source way .