cover news reporter Bian Xue
Article introduction:
1.ChatGPT's implementation means that it is a feasible technical route for machines to understand human language. At present, our country has also done a lot of work on large models of natural language understanding. The development of domestic related technologies is synchronized with foreign countries, and there is not much gap.
2.ChatGPT seems to be a very authoritative answer, but it is probably wrong. ChatGPT is a large language model trained based on massive data. The answers it can give in its Q&A are based on the training data. Therefore, as long as there are no questions in its training data, it is impossible to give accurate answers.
3. How to ensure that the future brought to us by artificial intelligence is friendly? The goal of the development of intelligent technology is to solve complex manual labor for humans. The development of AI technology should target human needs and solve practical problems in order to be friendly in the future. The
intelligence test result was 83 points. In the SAT test, I scored 1107 points (the median level of American students). I programmed a program, wrote a paper, and was able to make loofah and egg soup in a decent way.
Even for people who are not interested in artificial intelligence, it is difficult not to notice ChatGPT, a large-scale artificial intelligence robot developed by the American artificial intelligence research company OpenAI. Since its launch on December 1, ChatGPT has gone from 0 to 1 million users in just 5 days, and the intelligent dialogue system has been endowed with various meanings. Although OpenAI has publicly warned many times that ChatGPT "occasionally may produce incorrect or misleading information" and users need to use it with caution, it still cannot stop users from coming up with all kinds of questions to "make things difficult" for ChatGPT.
In the field of artificial intelligence, the birth of ChatGPT has revolutionized the universal concept of chat robots. In the field of general artificial intelligence, ChatGPT is the "first milestone". Four weeks after its launch, the heated discussion about ChatGPT in the technology community quickly cooled down, and the initial uproar has gradually become calm.
Search changes for the "chatGPT" term on the Baidu index in the past 30 days. (Image source: Baidu Index)
Has ChatGPT, which has been named a "subversive", "extinct"?
ChatGPT after "cooling down" is still a milestone in the evolution of AI
Computer scientist Alan Turing (Alan Turing) proposed it in 1950 as a way to measure intelligence: When humans talk to humans and computers, can they tell which is which?
In 2022, ChatGPT responded. Through ChatGPT, users can input questions in any natural language, and ChatGPT will give conversational answers. Although it may be a bit stiff, ChatGPT will remember the content of each conversation with the user, and combine previous questions raised by the same user with a large amount of existing information on the Internet to reply to the user's next question.
The social network Twitter is filled with users of ChatGPT, fully demonstrating the capabilities of artificial intelligence in generating artistic prompts and writing code.
CNET writer David Lumb listed some helpful ways ChatGPT can help with his writing; a doctor tweeted that he used ChatGPT answers to convince a health insurance company to pay for a patient's surgery. Zhang Li, who has used ChatGPT many times, told the cover news reporter: "I also tried to enter some recipes, such as braised pork, tomato and egg soup, and ChatGPT also gave inaccurate but correct methods."
Zhang Li used "chatGPT" to ask questions. (Picture source: Photo provided by the interviewee)
"The form of human-computer interaction currently achieved by ChatGPT has taken the interaction between humans and machines to a higher level." Associate Professor Yang Chunming, head of software engineering at the School of Computer Science and Technology of Southwest University of Science and Technology, CCF (China Computer Society ) and CAAI (China Artificial Intelligence Society), said in an interview with a cover news reporter that ChatGPT is based on OpenAI Built on the GPT-3 architecture launched in 2020, this large model based on large-scale data training also provides new ideas and approaches for research on artificial intelligence applications in specific fields. It is expected to solve human-computer interaction problems in some specific fields, such as some common sense and transactional questions and answers.
"In the context of the rapid growth of data scale and rapid improvement of computing power, it is inevitable that ChatGPT can achieve the current results and attention, especially the next generation GPT-4 may have better performance." Yang Chunming said that at present, ChatGPT has only achieved better results than the original dialogue robot in terms of multi-round chat dialogue and text generation between humans and machines, but this shows that this is a feasible technical route for machines to understand human language.
According to Yang Chunming, my country has also done a lot of work on large models of natural language understanding, such as the world's first knowledge-enhanced 100-billion-dollar model jointly developed by Pengcheng Laboratory and Baidu - Pengcheng-Baidu Wenxin (model version number: ERNIE 3.0 Titan). The model parameter scale reaches 260 billion, making it the world's largest single Chinese model. "Wudao, Pangu α, M6, etc. are all. The development of related domestic technologies is synchronized with foreign countries, and there is not much gap."
The "Wudao 2.0" series of models released by Beijing Zhiyuan Artificial Intelligence Research Institute in June 2021 launched China's first world's largest bilingual multi-modal pre-training model, with a scale of 1.75 trillion parameters, surpassing the Switch previously released by Google. Transformer, "has won the first place in 9 world-recognized Benchmarks, reaching the level of accuracy.
In June 2021, Beijing Zhiyuan Artificial Intelligence Research Institute released the "Enlightenment 2.0" series of models. (Picture source: Internet)
Seems very authoritative Most likely wrong
As an artificial intelligence product, ChatGPT is trained to recognize patterns in large amounts of text taken from the internet, and then further trained with the help of humans to provide more useful and better conversations. It is worth noting that the answers users get through ChatGPT may only sound credible and look authoritative, but they may very well be completely wrong.
There is no shortage of complaints from netizens. For questions that do not have clear answers, ChatGPT usually gives very vague answers. Mike Kraus, director of data science at artificial intelligence company Beyond Limits Krause once said: "If you ask a well-structured question and hope it gives you the right answer, you may get the right answer, and ChatGPT will answer it so well that it will look like it was written by a professor at Harvard University. But if you change the structure of the question, ChatGPT will make nonsense."
Software developer site StackOverflow banned ChatGPT from answering programming questions. The site's administrators warned: "ChatGPT's average rate of getting correct answers is too low, so posting answers created by ChatGPT is harmful to both the site and the users who are asking or looking for the correct answer." "
" ChatGPT is a large language model trained based on massive data. The answers it can give in its Q&A are based on the training data. Therefore, as long as there are no questions in its training data, it is impossible to give accurate answers. This kind of real-time update of training data will bring certain problems to the practicality of ChatGPT. Some questions with high real-time requirements, such as news, stocks, , etc., may not necessarily be able to obtain accurate answers. "
At present, ChatGPT is only aimed at attempts in human-machine dialogue. It answers some common sense questions and is more suitable for fields that do not require high professional knowledge. You can try to answer questions. Yang Chunming told the cover news reporter: "Although ChatGPT's current answers cannot avoid some factual errors, it is possible to solve this problem in the future. For example, training data can be updated in real time through online training, or knowledge reasoning or prediction methods can be combined with domain problems to solve certain problems with high real-time requirements. " Behind the rise of
, how to maintain intelligence and be good?
The philosopher Hume once said: "All science is related to human nature, and the study of human nature should be the foundation of all science. "Any science is more or less related to human nature. No matter how far away from human nature the disciplines seem to be, they will eventually return to human nature in some way.
As the core driving force of a new round of scientific and technological revolution, artificial intelligence, which was born in 1956, is profoundly changing the world. Over the long years, benefiting mankind has been the original intention of its technological development that has not changed."There will always be a risk in trying to develop friendly artificial intelligence technology, that is, we may create things that worry us." OpenAI co-founder Musk once mentioned his original intention: "The best barrier may be to allow more people to have access to and possess artificial intelligence technology as much as possible. If everyone can take advantage of artificial intelligence technology, there will be no possibility that a small group of people will have dangerous consequences due to having too powerful artificial intelligence technology alone.”
On December 1, local time, OpenAI released chatGPT on the social platform Twitter. (Picture source: Website screenshot)
The potential of artificial intelligence is undoubtedly huge, but how to maintain intelligence for good in the future digital world? Yang Chunming told the cover news reporter: "The goal of the development of intelligent technology is to solve complex physical labor for humans. The development of AI technology should target human needs and solve practical problems, so that it can be friendly in the future."
Although there is still some distance from a mature product, ChatGPT has taken the first step. In the near future, what other AI technologies will appear in our daily lives? ChatGPT cannot answer the question
accurately.
[If you have news clues, please report them to us. Once adopted, we will be compensated. Follow us on WeChat for reporting materials: ihxdsb, QQ for reporting materials: 3386405712】
