Jin Lei Alex from The Temple of Consort
Qubits | Official account QbitAI
"2 are plagiarism and 4 are irregular citations."
"The relevant responsible persons have voluntarily left their jobs."
"Restructuring of the Big Model Research Center."
……
A review report called A Roadmap for Big Model, jointly completed by 100 people, has attracted a lot of attention in academic circles at home and abroad because of suspected plagiarism.
In this regard, the beijing zhiyuan artificial intelligence research institute involved in the case reported the latest results after the investigation of the incident.
In addition to the points just mentioned, the circular also pointed out that the main organizational responsible person responsible for this "plagiarism gate" incident is the personnel of the Zhiyuan Big Model Research Center.
So what happened behind the academic plagiarism incident that has aroused heated discussion at home and abroad?
2 plagiarisms and 4 irregular citations
The focus of the "plagiarism gate", that is, this review report was led by the Great Model Research Center of Zhiyuan Research Institute, and invited a total of 100 researchers from 19 institutions at home and abroad to write.
According to the introduction of the circular, the review consists of a total of 16 independent monographs.
With the exception of article 12, each article has its own corresponding writer and corresponding author.
A core problem lies in the process of "collaboration".
In the "Identification of Responsibility for Organizational Negligence" in the circular, Zhiyuan Research Institute said:
The staff of the Zhiyuan Big Model Research Center did not strictly follow the process of academic publishing specifications, and uploaded the review report to arXiv on March 26, 2022 without confirmation with other authors, with primary organizational responsibility.
In addition, this briefing also responded to the previous netizens' doubts about the content of the review.
First of all, the Zhiyuan Research Institute admitted that there were 2 questionable fragments that belonged to plagiarism.
The first plagiarized fragment is section 2.3.1 of the second article of the review report, with multiple sentence repetitions totaling 179 words.
According to the "Types of Academic Misconduct by Authors" in the "Academic Publishing Code: Definition of Academic Misconduct in Journals":
Plagiarism: Paragraphs use textual expressions from other people's published documents, and although quoted, they do not quote the text used, or do not change the font, or do not use a specific arrangement.
This plagiarism fragment falls into this type of regulation and also meets level 5 of the IEEE Publications Services and Products Committee Operations Manual, "Guidelines for Determining Different Levels of Plagiarism."
The main point of the judgment is: a verbatim reproduction of the main part of an article, although quoted but lacking in clear distinction.
(Note: Plagiarism is divided into 5 levels, the 1st level is the most serious, and the 5th level is the most minor)
In this regard, the circular said that the two authors in this article (both of whom are members of the Zhiyuan Big Model Research Center) bear direct responsibility and negligence respectively.
The second plagiarized fragment, section 8.3.1 of the 8th article, has a 74-word repetition of the entire sentence.
The author in charge of this paragraph is also from the Zhiyuan Big Model Research Center, and the article was not confirmed with the corresponding author before publication, and should be directly responsible.
In this regard, Zhiyuan Research Institute said:
The above two authors have apologized to the original author in accordance with the corresponding corrective measures in the IEEE manual, and have received the understanding of the original author and fulfilled the relevant academic responsibilities they should bear.
In addition, the circular also said that 4 of the questioning fragments were reference irregularities.
They are:
- Article 10: There are a few repetitive texts
- Section 12: Section 12.2.3 has a total of 36 words of repetition, with no whole sentence repetition
- Section 14: Section 14.2.2 repeats multiple sentences and 63 words at a time
- Article 16: Section 16.1 there is multiple repetitions at a time
And the Zhiyuan Research Institute also made a report on the authors corresponding to the above problems.
As for the remaining questionable fragments, the circular said it was "a canonical reference" .
"All relevant responsible persons have left"
At the end of the briefing, the Zhiyuan Research Institute also announced the "treatment and rectification results".
First of all, the Zhiyuan Research Institute asked the authors of the articles that may have problems to apologize in writing to the original authors and said that "they have received feedback and understanding from the original authors."
Secondly, since the personnel involved in the "2 plagiarisms" are all from the Zhiyuan Research Institute's Big Model Research Center, the circular decides:
Restructuring of the division.
And the relevant responsible persons have voluntarily left their jobs.
As for the loopholes in the publication process, the Zhiyuan Research Institute said that it has "rectified the paper publication process and revised and improved the system of scientific research integrity and study style construction".
At the same time, in view of this incident, Zhiyuan Research Institute also made follow-up related plans:
The Institute plans to work with the academic community and industry to develop more rigorous citation specifications for literature and develop open source detection tools and systems for papers and code to avoid similar problems.
Event review
After reading the latest briefing, let's also review the beginning and end of this matter.
On April 8, Google Brain researcher Nicholas Carlini posted:
The A Roadmap for Big Model article copied a paper he had recently published.
He listed 10 paragraphs that were more obviously copied and marked in green the parts of the text that were consistent.
Carlini's experience of finding this situation is also quite dramatic: one of his dissertation collaborators originally wanted to see what zhiyuan's article was worth learning, but some parts became stranger and stranger the more they read it, and finally found that some paragraphs of the original paper directly copied their own article.
(Eat melon and eat your own head .jpg)
At the same time, through preliminary verification of collected data, Carlini found that the paper was also suspected of plagiarizing more than a dozen papers from other authors.
After Carlini exposed the suspected plagiarism of the Zhiyuan article, it quickly attracted widespread attention at home and abroad.
Even LeCun commented:
Some netizens believe that this is written by a team of students, who may not be clear enough about academic norms, while most of the others are pseudonyms.
Others shared academic misconduct they had encountered.
An author of a suspected paper came out and broke the news that the time left for his own writing at that time was very urgent, only more than a week, and he guessed that the time left for students may be less, so there was a large copy.
If you want to solve the problem, you should give the authors more time and more opportunities.
As things fermented, Zhiyuan officials also responded quickly.
On April 13, Zhiyuan issued a "Letter of Apology for the Problems of the "A Roadmap for Big Model" Roundup Report"; Two days later, it was announced that an investigation team would be set up to conduct an independent investigation into the papers involved.
On July 15, Zhiyuan officially announced the results of the investigation.
The investigation mechanism of Zhiyuan this time, and the final announcement of this open and transparent result, can also be regarded as a more appropriate way to deal with it.
Carlini, who initially broke the news that zhiyuan papers were suspected of plagiarism, later said that the article received much more attention than he expected, so he implored everyone not to turn this into a "witch hunt".
Carlini objected to the assertion that "all those involved in such matters should be immediately dismissed, should be banned from arXiv, etc.", and he advised you not to be too harsh on the authors of this article, but to focus more on the ethos of the academic field as a whole.
Also as Zhiyuan said in this notice:
Except for the relevant responsible persons of the Zhiyuan Research Institute mentioned above, all other authors in the review report did not plagiarize and engage in academic misconduct. We sincerely apologize for the negative impact and distress caused by this incident to these authors!
Once again, I sincerely thank friends from all walks of life for their supervision and criticism of Zhiyuan Research Institute!
— Ends —
Qubit QbitAI · Headline signing
Follow us and be the first to know about cutting-edge technology developments