Gpt-3 is a peer review online when it has been submitted for its own research

Pine From the Aofei temple
qubits | official account QbitAI

Why ？ There is something wrong with this paper , The first author doesn't look like a human name ：

The signature of yizuo GPT-3, Affiliated unit OpenAI.

Yes ！ You read that right ,GPT-3 I wrote a paper about myself .

before ,GPT-3 Have written news reports , Posted on the Forum , Even wrote novels and books .

I use GPT-3 stay 24 I wrote a book within hours —— And it has been published .

It seems that only academic papers are left , There has been no relevant attempt .

therefore , Swedish researcher from the University of Gothenburg Almira Osmanovic Thunström I'm going to try , The theme is not as good as a doll , let GPT-3 Write it yourself .

Write an essay 500 Word about GPT-3 Academic papers of , Scientific references and citations are added to the text .

At first, she didn't expect much , After all, the task prompt is still very vague , but GPT-3 She was stunned by the results given .

GPT-3 The academic language of is very standard 、 The references cited are also context sensitive .

Come here , She decided to produce a complete paper , And GPT-3 In the name of ······

GPT-3: I write papers for myself

In the official edition of the paper ,Almira Separate tips are designed for each chapter .

First It's a summary , First, let GPT-3 For the last generated 500 A word essay 200 A word summary .

Next Yes, introduce them separately 、 Method 、 Expansion of the results and discussion section .

It is worth noting that , When expanding this part of the method , Tips in addition to giving simple instructions , The corresponding determiners and requirements for writing are also added （ Don't show the exact value ）.

Last , The whole paper GPT-3 In just two hours .

We know GPT-3 You can write academic papers with yourself as the research object , And the efficiency is quite high .

But why should it write its own thesis ？

According to the researchers Almira The readme of , There are two main reasons ：

firstly , at present GPT-3 It's relatively new , There is little research on it , It can be used to test GPT-3 The accuracy of analysis when the amount of data is small ;

second ,AI Mistakes are not uncommon , What we want to do is not to publish AI Generated error messages , But this mistake will be part of the experimental order in the process of writing the paper .

GPT-3 Make mistakes , But it does not conflict with writing papers related to yourself .

When you finish writing your paper and are ready to submit , Some interesting things also happened ：

stay Almira In the readme of , The first problem encountered when submitting is ：GPT-3 What is your last name ？ What should I fill in its mailing address ？

reasoning , Last , Fill in the last name without , The mailing address and email are filled in Almira And his assistant .

This is only an insignificant part . Next, these may involve the law .

The first problem is ： Do all the authors agree to publish ？

This makes Almira A surprise ,GPT-3 It's not human ！

But I still summoned up the courage to ask GPT-3, Fortunately, its answer is yes .

Then there is another problem ：“ Are there any conflicts of interest among the authors ？”

GTP-3 towards Almira Promise not to .

therefore ,Almira Just put GPT-3 The papers written are uploaded to the French international preprint server HAL On .

Writing a paper is only the first step , If this paper is published , There are a lot of problems waiting to be solved .

GPT-3 Is it a tool or a partner ？

Not long ago ,Scientific American Posted a tweet introducing the story , And commented that ：

AI As the first author of this paper, many ethical issues have arisen —— This may overturn the subsequent publishing process .

stay twitter On , The issue of the author's signature has also triggered some discussion ：

A very good 、 Articles that challenge the current academic publishing model .AI Can I be an author ？ If possible , How to mark ？

Why? GPT-3 Not the corresponding author ?

If it can be published , Then these problems will follow ：

Should we consider AI Author identity ？

How should journal editors review other people's articles later , Should everyone be required to prove that they have not used GPT-3 Or the help of other algorithms ？
······

These problems have not been settled yet .

Rather than （GPT-3） It's a partner , It's more a tool .

But whether it will GPT-3 As a partner or a tool ？ These problems are inevitable .

Put aside these pending ethical issues , We can see from this study AI The development of ：

On a technical level , Netizens are optimistic about this research as a whole .

I hope this will bring some projects , Help researchers translate their research results into easier 、 faster 、 Better paper or paper form .

This is the beginning of the AI revolution . First , It can finish the paper in two hours , This is beyond the reach of human beings . The next step is Skynet .

at present , HAL Already given GPT-3 The author of this paper was assigned an editor , In the state of pending peer review .

If you're interested , A link to the original paper is attached below .

Original paper ：
https://hal.archives-ouvertes.fr/hal-03701250v1

Reference link ：
[1] https://www.scientificamerican.com/article/we-asked-gpt-3-to-write-an-academic-paper-about-itself-then-we-tried-to-get-it-published/
[2] https://www.reddit.com/r/Futurology/comments/vpoopq/we_asked_gpt3_to_write_an_academic_paper_about/
[3] https://twitter.com/sciam/status/1542474428428214272/retweets/with_comments