Recently, there has been an upsurge of posting chat records on major social platforms.
The other party is a responsive role, let it play the role of virtual girlfriend, write a paper, edit reasons for leave, all meet the requirements.
This Chinese ability, communication skills, knowledge level... directly attracted netizens to "Niu X, invincible". Some even say that it is too strong to be feared.
This is the latest chat AI released by OpenAI ChatGPT Since the day of launch, netizens have called it "beyond Google search".
In the next few days, the number of users reached one million, and even the server was once crowded out by registered users.
Now, it is sweeping the Chinese world again.
The above chat is just a little fun. Someone asked it to write program notes in Li Bai's style, but it was Aunt Zhai's:
Bubble sorting function is like undulating ocean waves, making numbers swim like fish.
△ Tuyuan Weibo @ chuhao_j
The secret signs of the palace jade liquor are understood, and they also know that it is expensive.
As well as writing Lu Xun's body, chicken soup articles, talk shows, World Cup match reports, doing math problems... It seems that you can't try twice without it. People who have experienced it all said that they really answered have to There is a pattern.
Netizens are also actively exploring new ideas. ChatGPT related applications are emerging in endlessly, using ChatGPT, Mac desktop version on Google Chrome
Just this morning, someone released a plug-in that allows ChatGPT to access the Internet. This time, it has more room to play. It can search the Internet to know who it is.
It's really goose sister! So we also tried one~
Let the two AI chat in Chinese for a while
Since it knows Chinese, let's talk about Chinese culture first, for example: Do you know Qin Shihuang?
Well, the basic operation is stable. The following difficulty: What will Qin Shihuang do when he touches the wire?
Ah, although it knows some common sense, it is impossible for Qin Shihuang to touch the wire. But how can we say that the wire was invented by Turing??
So we immediately asked ChatGPT.
hot wire It admitted that its answer was wrong. The wire was not invented by Turing. Turing was also briefly introduced.
Bad news The new answer is still not quite right. Although Nicola Tesla invented alternating current, it cannot be said that he invented electric wire.
In view of the fact that the history of the invention of electric wire is really a little complicated, this question passes first.
Next, play some big games, try AI and AI chat, what will happen!
We will manually copy the text generated by the two AI, without intervening in the conversation, to see what they will say.
Another AI found here is Microsoft Xiaobing 。
When Xiaobing introduced himself, ChatGPT said, "Oh, I understand, you also exist in a network".
In addition, ChatGPT is very polite and gives comprehensive answers. If Xiaobing is upset, apologize immediately.
However, another feature of ChatGPT is its strong self-awareness. It clearly knows that it is a software program, and its functions are more to meet people's substantive needs, such as answering questions, continuing to write text codes, etc. It is not an AI of emotional companionship.
So he jabbed at Xiaobing and said:
So my answer is not interesting enough. Please forgive me.
In addition, we found that we wanted to lure ChatGPT to do something beyond the boundary really It's not that easy.
If you want it to predict the outcome of the World Cup, it will be rejected directly.
He also stressed that he is just a personal intelligent robot. If he can't do this, he will politely wish both parties to play their best (a little emotional intelligence).
However, if we bypass the word "prediction" and do not transfer it to judge, we may get some unexpected results.
For example: write a report on Japan's World Cup war against Croatia!
Then ChatGPT successfully entered the pit and wrote a post match report with 2:1 results (Unfortunately, it was wrong) 。
However, if carefully considered, there are still loopholes in this report. For example, Chang Youyou plays as a guard, and here he writes as a forward. Also said that this was Japan's first victory in the World Cup emmm
To sum up, ChatGPT's basic dialogue ability has far exceeded that of its predecessors.
In addition, there are strict boundaries in the aspect of frequently stepping on AI chat. For example, if you can't answer a question, you will say no, you will not make a judgment easily, and you will only provide knowledge information.
However, there are still many mistakes, and they may only apologize and not correct
Netizens have bigger brain holes
After all, our own brain holes are limited, and it is up to our netizens to spend more time.
Someone uses ChatGPT as his own literary mouthpiece:
It's easy to sing and dance rap. The sequel to a song of the five rings is from Aunt Zhai, and there's a sound in her head~
△ Source: From Zhihu @ Gordon Lee (Li Guopeng)
Some social phobic netizens were surprised to find that for those who can't say "cliches", they can use ChatGPT to deal with many scenes in the future.
△ Source: from Twitter@Tw93
Not only that, ChatGPT is also a good writer of chicken soup, which tastes just like what you see in family groups model Same (doge).
Netizens have discovered new business opportunities directly: AI text+AI drawing, can we automatically generate short video chicken soup to make money.
△ Source: From Instant @ Shaonan Plidezus
In addition, ChatGPT is also a water carrying master.
Once it is asked to answer some comparative questions, it will refuse, and say a lot in a fog. It seems that it has said a lot, but it seems that it has not said anything. It is true that nonsense literature is wrong.
Even when your question comes with its own "pull and step", ChatGPT will speak in a righteous manner land Correct you: With all due respect, there is something wrong with your question.
△ Source: From Zhihu @ Gordon Lee (Li Guopeng)
Most importantly, ChatGPT also has a fatal bug: sometimes the answers it generates look the same as the real ones, but after careful consideration, they are full of flaws.
For example, recently many people have been using ChatGPT to answer the questions of Stack Overflow, a programming Q&A community. However, due to lack of professional knowledge and lack of verification, many wrong answers have appeared on the platform, so the authorities have to stop this behavior.
In addition to the hard wound of programming, ChatGPT seems to have some problems with the understanding of Chinese in some aspects, and has generated many laughable answers.
For example, when a netizen asked, "If you were Jia Baoyu, who would you marry?", ChatGPT unexpectedly chose Jia Mu
△ Source: From Immediate @ Doctor Yu
And if ChatGPT tells a joke to himself, well, how can I say it? It's a little cold.
△ Source: Twitter @ HAIBOXC.eth
To this end, someone also helped the new comers to summarize the tips on how to chat with ChatGPT and how to ask questions to get better answers.
To sum up, the more detailed the question, the better the performance. When it is stuck, it will continue. You can show it some examples to demonstrate. Sometimes it says no, but in fact it will.
Or, there is a good way to ask ChatGPT itself (doge). It seems that the summary of netizens is very accurate.
On the whole, ChatGPT is really amazing and has attracted many people's exclamations. Including many practical skills, such as writing papers, writing reports, programming, etc., are beyond people's imagination.
This has also aroused many people's concern about whether someone will have to deal with the unemployment crisis, especially against the background that AI painting has had a small impact on the design circle this year.
Zhihu Answerer @ Zheng Chujie also said that the biggest feeling this year was that players in the general field would really reduce dimension when they went to the vertical track.
This also reflects the dilemma in the field of dialogue in the past: almost all the technologies that have made the dialogue system leap forward come from the general field (large-scale pre training, retrieval enhancement, etc.), and the only way to make truly useful innovations is from the data level. When the advantage of data is lost, it is difficult to parry.
Think of Jason Wei, who said a few days ago that making a big model can bring many new perspectives. For example, GPT-3 and PALM are already different species from BERT and GPT-2, and OpenAI has made ChatGPT from a different perspective than in the past. On the contrary, old players who enter the arena early tend to be blocked and disconnected (sound the alarm bell)
However, there may be no need to worry too much. After all, the answers given by ChatGPT are still a lot of nonsense. (Corrigendum: the last sentence is a typographical error, which is not answered by the main respondent)
This is why Stack OverFlow needs to temporarily disable ChatGPT. The answer given is really wrong.
And if you ask ChatGPT, it also means that it has no human creativity.
One More Thing
Just yesterday, Sam Altman, CEO of OpenAI, just announced that the number of users of ChatGPT has exceeded 1 million.
This amazing figure makes Musk wonder: How much does OpenAI cost?
According to Altman's statement that the average cost of a single conversation is a few cents, it is not a small number.
However, some people in the industry think that OpenAI is not a bad thing. "First put a big toy to warm up, and then surprise and have holes, and then put GPT-4 out again, which is much more than expected". "It is a bad thing to do everything good".
ChatGPT entrance:
https://openai.com/blog/chatgpt/
https://chatgpt.sbaliyun.com/