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

The survey found that 89% of American college students already use ChatGPT to do their homework, and 72% of them also support the ban. In this regard, the attitude of the teachers is mixed, which is worth pondering.

ChatGPT has only been born for two months, but the "blockbuster" it has dropped on the world has never stopped.

Although in many schools, teachers guard against ChatGPT like a scourge, they still ban it again and again.

A survey shows that 89% of American college students now use ChatGPT to do their homework.

89% of the students use ChatGPT to do their homework. Yes, even, the true proportion is higher.

Although ChatGPT was completely banned by New York's education system, how could it be really banned with the ingenuity of the students?

Sure enough, now educators have to face the reality: students are already cheating unscrupulously with ChatGPT.

For example, when Antony Aumann, a professor of philosophy at the University of Northern Michigan, graded himself, he found that the first paper in his class was written in ChatGPT.

Study.com, an online course provider, conducted a survey of 1000 students over the age of 18 and asked them about their use of ChatGPT in class.

The results show that--

More than 9 out of every 10 students know about ChatGPT, far more than primary school educators

More than 89% of students use ChatGPT to finish their homework

48% of the students completed quizzes with ChatGPT, 53% used ChatGPT to write papers, and 22% used ChatGPT to generate outline papers.

Surprisingly, 72% of college students think that ChatGPT should be banned in the university network.

These astonishingly high data make us have to face up to this phenomenon: AI has been integrated into the social structure of human beings, and has produced extensive and far-reaching consequences.

Interestingly, although nearly 90% of students use ChatGPT to do their homework at home, nearly 3/4 of students want ChatGPT to be banned at school.

In other words, do not suffer from scarcity but inequality, no matter in which scene, students want to stand on the same starting line.

Either we all use it together or we don't use it at all.

What do the teachers think? At the same time, Study.com also surveyed more than 100 educators to learn more about how they feel about ChatGPT.

72% of university professors who know ChatGPT are worried about its impact on cheating, compared with 58% of primary school educators

More than 1/3 (34%) of educators think that ChatGPT should be banned in schools, while the remaining 66% support students' use.

It seems that the teachers are far more open-minded than we thought. So how are they going to use ChatGPT?

According to a survey by Study.com, 21% of teachers have begun to use ChatGPT to assist their teaching work.

7% use ChatGPT to provide writing tips

5% use ChatGPT to help teach

4% use ChatGPT to develop course plan

4% use ChatGPT to teach writing style

3% use ChatGPT as a digital mentor

It can be seen that, contrary to our impression, most of the teachers who participated in the survey had a relatively enlightened attitude towards AI, and 66% thought that ChatGPT could be used as a resource to help students.

In contrast, students themselves have much less trust, with up to 72% of people believing that ChatGPT should be banned in schools.

In the whole human history, the birth of a new thing is often accompanied by a lot of controversy.

Obviously, in the face of ChatGPT, a two-month-old "newborn", there is no consensus within the teaching team.

Some of the teachers have a very clear attitude towards students' use of ChatGPT-pure cheating!

Professor of New York University: avoid ChatGPT like a plague god! These days, many schools in the United States have opened, and there is no doubt that the hottest topic among teachers and students is ChatGPT.

At NYU, the "academic integrity" section of the syllabus explicitly treats the use of AI as cheating and forbids it.

In addition, the students received a warning from the professor on the first day of class.

In a class at the Tisch School of Art at New York University, the professor wrote bluntly on the syllabus--

"Q: is it considered cheating to use ChatGPT or other AI tools that generate text or content? A: yes. "

Even in classes where there is no need to write a paper, the professor gives a ChatGPT warning.

A macroeconomics syllabus reads: "We deliberately keep time tight, so you can't have time to consult books, ChatGPT or other resources and complete the test at the same time." Students are not allowed to communicate with anyone (including ChatGPT) during the 24 hours of the test. "

Of course, it is well known that ChatGPT is often stupid when dealing with math problems, so math professors do not have to worry about this.

Jenni Quilter, associate dean of the School of Arts and Sciences at New York University, says professors are now worried that students will use ChatGPT to cheat.

According to Quilter, there were incidents of students using ChatGPT as early as December.

"using ChatGPT without permission will have the same consequences as any academic plagiarism, including redoing homework, deducting points, and writing checks. "

David Levene, a classics professor at New York University, says he is keeping a close eye on all ChatGPT-related plagiarism.

"I have clearly warned students that it is cheating to use ChatGPT in any form unless I have my permission. "

"I also told them that I had tried to write papers in ChatGPT, and its best score was Bmuri and the worst was F. So if they want to get a higher score than B -, they should avoid it like a plague god! "

The fears of NYU professors are not unfounded.

According to a survey conducted by Stanford Daily, 17% of students have used ChatGPT to complete their homework and exams for the fall semester.

However, the proportion on the Stanford side is obviously much lower than that of 89% and 48% of Study.com.

Will AI make students "brain atrophy"? Many professors are worried that AI chatbots will have a disastrous impact on education.

"just because I have a machine that can help me lift the dumbbell doesn't mean my muscles will develop," Johann Neem, a history professor at the University of Western Washington, told the Wall Street Journal.

"similarly, having a machine that can write papers does not mean that my mind will develop. "

But other professors believe that ChatGPT's powerful technology should be used to prepare students for the new reality.

"I hope it gives you enough inspiration and education to make you want to learn how to use these tools, not just learn to cheat better," said Alex Lawrence, a professor at Webber State University. "

Ethan Mollick of the University of Pennsylvania says he hopes his literature students can use technology to "write more" and "write better."

"ChatGPT is the power multiplier of writing," Mollick added. "I want them to use it. "

Google can, but not ChatGPT? Although it has caused a storm of academic integrity, many experts believe that this technology is only the beginning of a new learning era-AI writing tools are the future of learning.

"I think this is a major moment for human capacity improvement," said Phillip Dawson, director of the Digital Research Center at Deakin University. "

"in my opinion, students who graduate in five years can do a lot more than students now, because they have these AI tools. "

He drew an analogy between the student who wrote the paper and the pilot who flew the modern plane. "Yes, you have to learn to use all the instruments, you need to know how they work, but you also need to fly the plane if the instruments fail. "

Dr. Cheryl Pope, a lecturer at the School of computer and Mathematics at the University of Adelaide, said that ChatGPT is very suitable for writing first drafts, but it cannot replace the need for manual editing and fact-checking. "you need to understand this topic before you can comment on the answers it produces. "

ChatGPT can give you a few steps, but it can't get you a high score. But its possibility is exciting and can lead us to higher standards. Just like we have different expectations for a two-hour written exam and a two-month paper.

Another reason is that getting help requires a lot of social resources.

It's a shame to ask someone a stupid question, but in the face of AI, we never have such a worry.

Stanford's strongest detector: DetectGPT has attack and defense, and the AI cheating detection tool, which can solve teachers' worries, has also been born quickly.

Recently, a team of researchers from Stanford University came up with a new method for detecting AI-generated text-DetectGPT.

To sum up:

DetectGPT detects whether the text comes from the pre-training language model by using the local curvature of the logarithmic probability function of the model (the region of negative curvature is often occupied by LLM).

DetectGPT only uses the logarithmic probability calculated by the interest model and the random disturbance from another general pre-training language model, such as T5, without training a separate classifier, collecting real or generated paragraph data sets, or watermarking the generated text.

The detection effect of DetectGPT is better than the existing zero sample (zero-shot) methods, especially the detection rate of false news generated by 20B parameter GPT-NeoX is improved from 0.81 AUROC to 0.95 AUROC.

Https://arxiv.org/ abs / 2301.11305 We can observe that the machine-generated text (left) tends to be located in the logarithmic negative curvature region, while the nearby samples have a lower logarithmic probability of the model on average.

In contrast, the human text (right) does not obviously occupy the region of negative logarithmic probability curvature.

Next, you want to determine whether a paragraph of text is generated by a specific LLM, such as GPT-3.

First of all, DetectGPT is required to use a general pre-training model (such as T5) to slightly disturb the paragraph. Then DetectGPT is asked to compare the logarithmic probability of the original sample with that of each disturbed sample.

If the average logarithm ratio is high, the sample is likely to come from the source model.

The specific test results are as follows:

In addition, the supervised detection model trained by large datasets of real and generated text performs the same or even better than DetectGPT on distributed text. (above)

However, for new areas, such as PubMed medical texts and German news data from WMT16, zero-shot is available out of the box, while supervised detection methods will collapse due to excessive distribution deviation. (below)

However, DetectGPT itself has obvious limitations.

First of all, DetectGPT is based on the white-box hypothesis, that is, we can evaluate the logarithmic probability of the model. For the models behind API (such as GPT-3), it costs money to assess probabilities.

Secondly, DetectGPT needs to obtain a reasonable disturbance function. Although the author uses ready-made masking models such as T5 and mT5 (for non-English languages) in this work, if the existing models do not represent space well, then the performance of DetectGPT in some areas may be degraded.

Finally, DetectGPT is more computational than other detection methods because it needs to sample and score the disturbance set of each candidate paragraph, rather than just scoring the candidate paragraphs.

Although DetectGPT is not open at this stage, it is not a big problem.

After all, there are still many tools on the market that can be used directly.

GPTZero, in particular, is not only free, but also effective.

The editor's own test found that the latest version of GPTZero can even clearly point out which paragraphs in a text are generated by AI and which are written by humans.

Experience link: in the principle of https://gptzero.me/, GPTZero mainly depends on "confusion" (randomness of the text) and "sudden" (change of confusion) as indicators.

In each test, GPTZero will also pick out the sentence with the highest degree of confusion, that is, the most human-like sentence written by AI.

Reference:

Https://study.com/resources/perceptions-of-chatgpt-in-schools

Https://stanforddaily.com/2023/01/22/scores-of-stanford-students-used-chatgpt-on-final-exams-survey-suggests/

Https://www.vice.com/en/article/n7zxe7/nyu-professors-tell-their-students-do-not-use-chatgpt

Https://www.vice.com/en/article/xgyjm4/ai-writing-tools-like-chatgpt-are-the-future-of-learning-and-no-its-not-cheating

Https://arxiv.org/abs/2301.11305

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.