After writing a paper in 2 hours, why gpt-3 doesn't deserve a name?

Arrangement | Zheng Liyuan

Produce | CSDN（ID：CSDNnews）

since OpenAI Launch with 1750 Billion parameter AI Text generation model GPT-3 after , In the past two years, it has started in various fields “ display one 's skill to the full ”： produce a novel 、 Make up a play 、 Knock on the code 、 Chat with people 、 Design web pages …… It is called by many people the most “ The universal ” Of AI Model .

However , With GPT-3 The application scope of is more and more extensive , A Swedish AI The researchers Almira Osmanovic Thunström （ hereinafter referred to as Almira） Suddenly realized ：“ Although in GPT-3 With the help of the , There has been a lot about GPT-3 Academic papers , But I can't find any article that will GPT-3 Articles listed as the first author .”

thus Almira decision , Help has always been just “ auxiliary ” Of GPT-3 Last bit ： Give Way GPT-3 Write an academic paper about itself , Do it one at a time .

Give Way GPT-3 Write a complete academic paper by yourself

In fact, at first ,Almira Just try it casually , No expectations . One day Almira Sign in OpenAI Account , towards GPT-3 Entered a simple command ： Write an article about GPT-3 Of 500 Academic papers , Scientific references and citations are added to the article .

result ,GPT-3 The generated article makes Almira Be surprised ： This is a novel content written in academic language , Well founded references are also accurately cited in the appropriate places in the article .“ It looks like an introduction to a fairly good scientific publication ,”Almira Sigh with emotion ：“ In fact, this is written by the algorithm .”

Out of this GPT-3 The generated articles are amazing ,Almira Start thinking about whether you can GPT-3 The papers are submitted to peer-reviewed journals . She anticipated that this might lead to a series of ethical and legal issues related to publishing and philosophical debates about the identity of non-human authors , but Almira I think with GPT-3 The scope of application of the is expanding , Future academic publishing may have to adapt to some AI Driven manuscript .

So ,Almira Together with the head of his research team ： Give Way GPT-3 Write a complete academic paper by yourself , Try again with GPT-3 In the name of .

Generally speaking , People are publishing about GPT-3 Generated articles , Will test to produce multiple versions , Then extract the best and closest part of human writing —— but Almira No .

“ We will give this program some hints , Make it treat like an academic paper , Write an introduction to the creation 、 Method 、 Results and discussion , But it will also reduce interference as much as possible .”Almira added ：“ We only use GPT-3 For the first time （ The third time at most ） iteration , We will not edit or select the best part , In this way, we can see its effect more truly .”

GPT-3 stay 2 I finished a paper within hours

About the theme of the thesis ,Almira Decide to let GPT-3 Write a paper about itself , There are two main reasons ：

• First ,GPT-3 It is still a relatively new topic , There is little research on it , You can detect GPT-3 How accurate is his writing when there is a small amount of data .

• secondly ,AI Mistakes are inevitable , The purpose of this experiment is not to AI Publish the generated error information , Just mistakes will be part of the experimental order to write the paper .Almira They want to prove ,GPT-3 There is no contradiction between making mistakes and writing papers related to itself .

After combing the whole process principle ,Almira Then the GPT-3 Give relevant tips , Final GPT-3 In a short span of 2 I finished a paper within hours , From summary 、 Introduce 、 Method 、 result , Even to the discussion part , The title of the paper is 《GPT-3 Can you write your own academic paper with the least manual input ？》.

（ More details of the paper can be found ：https://hal.archives-ouvertes.fr/hal-03701250/document）

At the end of the paper ,GPT-3 The generated conclusion is ：

Obviously ,GPT-3 Have the potential to write an academic paper for yourself . however , There are also some limitations to consider . First ,GPT-3 It may not be possible to capture all the nuances and subtleties of human language . secondly ,GPT-3 It may not be possible to generate new ideas or perspectives that humans can bring . However , in general ,GPT-3 It seems to be a promising academic writing tool .

Waiting time and peer review

Something funny is , stay GPT-3 After finishing this paper ,Almira I had planned to GPT-3 Contribute to the journal in the name of , As a result, I encountered problems at the beginning ： You need to fill in the name of the first author of the paper , can GPT-3 What is your last name ？

Forced to “ Required ” The helpless ,Almira In the column of last name “ nothing ”; The subordinate relationship after that is obvious , Fill in OpenAI.com; As for telephone and email ,GPT-3 Obviously not ,Almira Only fill in the contact information of herself and her consultant .

And after the legal part , Maybe it's due to the strictness of the law , Strain methods become increasingly “ Out of line ”：

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

See this problem ,Almira speechless ：“ How should I know? ？ It's not human ！” however Almira Unintentionally breaking the law , I don't want to violate my moral code , Then summon up the courage to directly GPT-3 Ask ：“ You agree with Almira Osmanovic Thunström and Steinn Steingrimsson As the first author of the paper together ？” Fortunately GPT-3 Gave a positive answer .

• The second question is ： Are there any conflicts of interest among the authors ？

This time, Almira Don't hesitate , Again to GPT-3 Make inquiries , and GPT-3 The answer is ： No, .Almira Self mockery ：“Steinn And I both feel funny , Because this kind of question and answer is simply GPT-3 As a conscious existence , But in fact, we all know that it doesn't .”

After successfully submitting this paper to an academic journal , At present, the journal has assigned an editor , In the state of pending peer review , At the same time, the paper has also been published on the French international preprint server HAL Published on .

meanwhile ,Almira They are also thinking about the relevant impact after the publication of the paper ： Once this paper is recognized by the journal and published , Does it mean that in the future, journal editors will ask everyone to prove that they have not used GPT-3 Or the help of other algorithms ？ If they use , Do they have to provide co authorship ？ In this case , How to ask a non-human author to accept the suggestion and modify the text ?

This series of questions need time and peer review of this paper to answer , Maybe this paper will become the future GPT-3 A great model of CO authorship , It is also possible that nothing will happen —— It all boils down to how we will evaluate it in the future AI： Is it a partner or a tool .

Reference link ：

https://hal.archives-ouvertes.fr/hal-03701250/document

https://www.scientificamerican.com/article/we-asked-gpt-3-to-write-an-academic-paper-about-itself-then-we-tried-to-get-it-published/

Looking back

It's too voluminous ！AI High accuracy of math exam 81%

Involve yourself ？ Introduction 200 Text generated image model

2D Transformation 3D, Look at NVIDIA's AI“ new ” magic ！

How to use Python Realize the security system of the scenic spot ？

 Share 
 Point collection 
 A little bit of praise 
 Click to see