Thinking Space

ChatGPT is an artificial intelligence (AI) chat robot program developed by the American Artificial Intelligence Research Lab.After it was released on November 30, 2022, the number of users reached 100 million.The topics of ChatGPT and AI have always been hot, and people have worried about which work will be replaced.But after some understanding, people may temporarily let go of this concern, because what is imminent is not a question of how humans will be replaced, but what kind of challenges of this generation of AI to our reality and the present,And what requirement for the education system.

For more than two months, people have made a variety of chats with this AI with curiosity, and gradually discovered that although it can naturally answer various questions like humans, they can also generate it.The perfect structure of the structure is also significant.

A friend graduated from the Department of Computer Department of the National University of Singapore. He has cultivated for 167 years in the IT industry. Recently, he also tried to use Chaatgpt in programming.When he tried a new programming language Julia, he encountered a small problem that he could not quickly use Google search to get the answer, so he asked ChatGPT, but the solution it gave seems reasonable but cannot run really. Finally, friends still check through viewing.Julia built -in document, plus experience -based guess, solves the problem.Although ChatGPT did give detailed answers, friends described: "It looks like that, but various Zhang Guanli Dai belongs to a serious nonsense."

Recently, a Chinese netizen Xiao Wang also shared a conversation with ChatGPT online.One of the questions is "Can lyrics and humor never be fused with each other?" It replied: "No. Lyrics and humor may not necessarily integrate. In many literary works, lyrics and humor are often organically combined. Lyrics are lyric.It has its own unique charm and humor, but under the right circumstances, the combination of the two can enhance the effect of literary works. "Xiaowang asked it to give it an example, it listed a dialogue, and said that it was the Titanic Titanic number., Dialogue of the male and female protagonists.Netizens found that the conversation did not exist in the movie at all, and it was made up.

ChatGPT's popularity has not decreased, and Google launched ChatGPT's competitor Bard on February 6.But on the first show, it made a factual mistake.According to a GIF movement shared by Google, Bard was asked: "I can tell my nine -year -old children about the new discoveries of James Weber Space Telescope (JWST)?" Bard's answer includes: the telescope shot out of our solar system.The first photo of the planet.Many astronomers and celestial physicists pointed out on social media that this was incorrect. The first outer planetary image was taken with VLT/NACO in 2004 Chauvin et al. With VLT/NACO.

Google and OpenAI both build their AI on text generation software. Although these AI can say, they are also prone to fake the facts.Some AI industry commentators said: "They are not searching for the database of proven facts to answer questions, but receive a large number of text corpus training and analyze the mode to predetermine the probability of the next word in any given sentence. In other words, in other words, in other words, in other wordsThey are probably probably, not certainty. "

Others pointed out: "This idea can only learn the statistics of the language itself, and there is no way to obtain the logic behind the language. Therefore,‘ a serious nonsense '.

This is also a lifelong professor at New York University and chief AI scientist at Meta, Yang Likun, does not think that ChatGPT will have major innovation, nor does it think that the main reason why the large language model is the correct direction of GM AI.

When someone asks ChatGPT similar to this question: "Is the information you provide accurate?" "How do you judge your answer is correct?"Technology, by learning a lot of text data, matching keywords and contexts, and then generating answers from known data. When answering questions, I will give the most correct answer based on the data and models I learned, but it cannot guarantee that each answer cannot be answered.It's correct. "

Chatgpt's answer is frank.This requires users to have certain information literacy and speculative ability, which can recognize which are "reliable", which are no factual basis, which exist in reality, and which are completely made up of AI.How people find the correct information they need in the ocean of information in the ocean of information, which requires both a systematic knowledge system and speculative ability.This speculation should also be the focus of education.

In fact, many netizens have found that the answer of ChatGPT is not necessarily correct. It is shocking to imitate experts and manipulate text.It also has an amazing ability that can learn during the dialogue and re -modify its answer according to the user's feedback.Although the aforementioned senior IT friends are disappointed with the analysis and reasoning ability of ChatGPT programming, it also acknowledges that its text tissue ability is very good, and "intelligence" still exceeds a lot of humans who "speak without the brain".He is also very optimistic about the combination of AI and search engines.He believes that if AI combined with Internet search, the source link to the source will help people judge the authenticity.Such a deep -combined tool will greatly improve people's work efficiency.

People also find that ChatGPT will try to make themselves seem "objective and neutral" when answering sensitive questions in politics, economy, religion, culture, etc., and the answer is generally quite "political correct".This also shows the results of artificial intervention and correction during the learning process.However, if students only stop at these "political correctness" answers and do not do in -depth thinking and identification, they will easily lose their ability to deal with sensitive issues.

The ultimate copy of Chinese writer Li Zhiyong: How artificial intelligence will promote the book of social changes, and there is a chapter discussing "the risk of discharge of artificial intelligence".He said: "Artificial intelligence does not give a completely certain result. The behavior under its guidance is a probability, which makes the risk of out of control that the system that rely on artificial intelligence."Practice "and" data pollution ".

When you need to make a value judgment and make decisions, the machine will not be like a person. When weighing the advantages and disadvantages, you can grasp the size, but it is easier to bias aside.In fact, humans themselves may not be able to grasp the scores, because it involves many philosophy and ethical issues.No matter what you choose, humans have to face and bear the consequences of their choice, but AI machines do not have to.

As for "data pollution", ChatGPT can be said to have solved this problem to a certain extent, otherwise it cannot be so "political correct".

In 2016, Microsoft released a chat robot named TAY, but within a day, it began to make improper remarks, such as hate women's remarks, also applauding the Nazis, saying that the massacre was fabricated, etc.It was forced to go offline soon.Microsoft explained afterwards that this is a learning machine, because the bad induction of some people who communicate with it will have some improper reactions.

Another hidden worry is that we cannot know exactly what data to enter for such AI machines, and there is no way to accurately determine what kind of behavior they will have.predict.Li Zhiyong believes that to hedge these risks, what is needed is the network that can be connected and scattered and efficient.

Perhaps because of the above reasons, ChatGPT will answer "AI Ethics" when answering questions similar to "what research directions will be popular in the future".Its rise also brings us a lot of thinking: not only the application prospects of AI, but also the choice of technical ethics and value.So, educationIt is also necessary to emphasize respect for multiculturalism and diversified values, and teach how to use empathy, bridges different differences, seek common ground while depositing, and improve the ability to deal with complex and sensitive issues.The more advanced AI, the higher the human civilization needs to face the impact and impact it brings.

The author is the creator of the children's book, the former media person