Shulou(Shulou.com)11/24 Report--

Recently, an associate professor at Cornell University investigated the attitude of the industry in writing papers with ChatGPT. The attitudes of those who support writing papers with ChatGPT are roughly the same as those of their opponents, but the longer they stay in the scientific research circle, the more they oppose ChatGPT.

After the release of ChatGPT, whether "using language models to assist thesis writing" is cheating, plagiarism or reasonable use of writing tools, there is still no unified standard in colleges and universities and paper conferences, and even different researchers hold different attitudes towards ChatGPT.

Some people think that ChatGPT can improve the quality of writing and reduce grammatical errors for non-native English speakers, while others think that the generated text does not belong to the author and violates academic morality.

Recently, while attending ICLR 2023, Alexander Rush, an associate professor at Cornell University, interviewed a number of attendees about their attitudes towards "writing papers in LLM".

As a result, the proportion of support and opposition is basically 50-50, and the degree of support and opposition is about the same.

Should ICLR prohibit LLM from generating papers? However, if we divide it according to the population of master's degree, doctor's degree and post-doctorate, we can see that the number of years of education is increasing, and the disapproval rate is also on the rise, among which the master's degree students are more inclined to use it, but there are also great differences in using language models.

Researchers who have published more than 10 papers most support the use of language models to write papers, but there are also great differences; in contrast, novices who have published only one paper and non-native English speakers and native English speakers generally support using language models to write papers, but native English speakers are more divided.

During the research, Alexander Rush also asked, "should ICLR disable LLM to generate reviews?" "

As you can see, compared with using LLM as a tool when writing papers, more than 60% of researchers believe that LLM should not be used to review manuscripts.

And those who have worked longer in the scientific research community, such as doctorates, postdocs and researchers who have published multiple papers, are also more likely to ban LLM review.

In terms of the use of LLM to assist paper writing, we can see that the most common scenario is grammar correction, but most people still use LLM occasionally or intentionally to modify grammar.

In the related work research, Latex chart production, the outline of the paper, the number of researchers using LLM as a daily tool has been significantly reduced, but many people have expressed interest in exploring the use of LLM in these areas.

Finally, there are the views of mentors and professors on banning LLM writing: most people's view is that there are pros and cons, but generally speaking, professors in the industry are more inclined to ban LLM,4% and believe that "restricting the use of LLM may be effective." 10% of professors believe that "although it should be banned, some people will still use it."

All in all, Rush concluded that most people still hold a wait-and-see attitude towards LLM as a scientific research tool, and that language models raise questions about "community trust", "cultural norms and biases" and "research and financial incentives". Policy issues are important.

Netizens commented on Turing Award winner Yann LeCun retweeted and said: as the co-founder and chairman of the ICLR Foundation, I am also a postdoctoral fellow and have published more than 10 papers, but I am very much in favor of letting authors use the tools they find most useful to help them write technical papers, including large language models.

Most netizens are in favor of using LLM for scientific research, believing that all researchers should master the ability of AI collaboration in the future.

Some netizens even joked that those who support language models are fake researchers, and real researchers never use spell checking or calculators, but carve papers directly into stone!

Some people think that ChatGPT is an artifact for writing Latex diagrams. It used to take hours to find the answer on stack exchange, but now it only takes a few minutes!

ChatGPT is also very helpful in training scientific research skills, such as using calculators as mentors to train your mental arithmetic skills.

And ChatGPT can discuss the theme of the paper with researchers, identify loopholes and improve the paper, which is a good thing. After all, not everyone has a talented partner to communicate with.

Do you think researchers should introduce LLM as a writing and review tool?

Reference:

Https://twitter.com/ylecun/status/1653752235514376192

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.