GPT-3 When I studied myself, I have submitted itself, and a peer review online squatting

Quantum position | Public account QBITAI

Huh? This paper is a bit wrong. The first author does not look like a human name:

The signature GPT-3 of the first works, Openai, which is affiliated.

Yes! You read it right, GPT-3 wrote an article about yourself.

Earlier, GPT-3 has written news reports, posted posts on the forum, and even wrote novels and wrote books.

I wrote a book within 24 hours with GPT-3-and have been published.

It seems that there are only related attempts in academic papers.

Therefore, Almira Osmanovic Thunström, a Swedish researcher from the University of Gothenburg, intends to try it. The theme is not as good as a baby, so let GPT-3 write itself.

Write a 500-word academic paper on GPT-3, and add scientific reference and citation in the text.

At the beginning, she didn't have a lot of expectations. After all, the task prompt was still very vague, but the result given by GPT-3 made her stunned.

GPT-3's academic terms are very standardized and referenced reference materials are also related to context.

At this point, she decided to make a complete paper and submit it in the name of GPT-3 ...

GPT-3: I write a paper to myself

In the formal version of the paper, Almira designed a separate prompt for each chapter.

The first is the summary, first let GPT-3 summarize a 200-character papers generated last time.

The following is the expansion of the introduction, methods, results, and discussion parts, respectively.

It is worth noting that when the method is expanded, it is prompted to add corresponding limited words and the requirements when writing (do not show exact values).

Finally, the entire paper GPT-3 was completed in just two hours.

We know that GPT-3 can write academic papers in the research object, and the efficiency is quite high.

But why do you want to write the paper on yourself?

According to researchers Almira, the main reason is the following two points:

First, the current GPT-3 is relatively new, and there are very few research on it. It can be used to test the accuracy of the analysis of GPT-3 under a small amount of data;

Second, it is not uncommon for AI to make mistakes. What we want to do is not to publish the error message generated by AI, but that this error will be part of the experimental command during the writing process.

GPT-3 can make mistakes, but it does not conflict with writing related papers.

When writing the paper to prepare for submission, some interesting things happened:

In Almira's self-report, the first question encountered during the submission is: what is the last name of GPT-3? What should its mailing address be filled in?

Think about it. Finally, the surname was filled in, and the mailing address and email were filled with Almira and its assistants.

This is just an irrelevant part. The following can involve the law.

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

This surprised Almira, GPT-3 is not humans!

But I still took the courage to ask GPT-3. Fortunately, its answer was yes.

Then there is another question: "Is there any conflict in the author?"

GTP-3 guarantees that there is no ALMIRA.

As a result, Almira uploaded the papers written by GPT-3 to the French International Pre-printing server HAL.

Writing the paper is only the first step. If this paper is published, there is a large series of problems waiting to be solved.

Is GPT-3 a tool or partner?

Not long ago, Scientific American posted a tweet to introduce the story and comment:

AI as the first author of the paper has caused many ethical issues -this may subvert the post -publishing procedures.

On Twitter, the issue of the signature of the thesis author has also triggered some discussions:

A very good articles challenging the current academic publishing model. Can AI be the author? If you can, how to mark it?

Why is GPT-3 not a communication author?

If you can publish, then these problems will follow:

Should I consider the authors of AI?

How to review other people's articles after the journal editor, should everyone ask everyone to prove that they have not used GPT-3 or other algorithms? ·······

These issues have not yet been concluded.

Rather than saying (GPT-3) is a partner, it is a tool.

But whether GPT-3 is regarded as a partner or a tool? These problems are inevitable.

Aside from these ethical issues to be treated, we can see the development of AI from this study:

At the technical level, netizens have a more optimistic attitude on this research.

I hope this can bring some projects to help researchers convert their research results into easier, faster and better papers or forms.

This is the beginning of the artificial intelligence revolution. First of all, it can complete the paper within two hours, which cannot be done by humans. The next step is to be Skynet.

At present, HAL has allocated editors for this paper written by GPT-3, and is in a state of evaluation of peers.If you are interested, the original thesis link is attached below.

Thesis Original: https://hal.archives- means.fr/hal-03701250v1

Reference link: [1] https://www.scientificAmerican.com/article/we-sked-gpt-3-write-an-Academic-about-ITSELF-we-To-Get-

- END -