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

AI summit ICML solicitation day is just around the corner, but the new policy on moral standards has aroused dissatisfaction among netizens! The rules require authors not to use large-scale language models, and netizens' comments area asks: why?

Yesterday, the International Machine Learning Conference (ICML) issued a draft announcement for 2023 papers.

The paper will be submitted from January 9 to January 26.

However, the requirements for a "code of ethics" in this meeting have aroused a lot of dissatisfaction.

LLM is out of control, so let's just drop ban. According to General Assembly policy, all authors and members of the procedural committee, including reviewers, should follow a standard code of ethics.

Any form of plagiarism and the unethical use of privileged information by reviewers, domain chairs (AC) and senior domain chairs (SAC) are strictly prohibited, such as sharing this information or using it for any purpose other than the review process.

Papers that contain text generated from large-scale language models (LLM), such as ChatGPT, are prohibited unless the generated text is presented as part of the paper's experimental analysis.

All suspected immoral acts will be investigated by the Ethics Committee, and individuals found to have violated the rules may face sanctions. This year, we will collect the names of individuals found to have violated these standards; if individuals representing conferences, journals, or other organizations request this list for decision-making purposes, we may provide them with this information.

Among them, the requirement of "prohibiting the use of large language models to write papers" has been hotly discussed by netizens.

This news has been released, netizens have commented on ICML Twitter: "Why can't we use a large language model?" "

AI papers cannot use AI, is it reasonable for Yann LeCun to forward and comment: "large language models cannot be used, meaning medium and small language models can also be used." "

"because spell checking applications and text prediction are also language models," he explained. "

Finally, LeCun Yin Yang said: "next year's ICML moral code should be: researchers must do all kinds of calculations with their bare hands, not with the help of computers; all kinds of graphics must be drawn by pen and paper; from today, the official language of ICML will become Swedish (Sweden:? ). "

Finally, I did not forget to forward a movie "Crazy" to sum up my comments on ICML policies.

Under the LeCun post, netizens showed their magical powers and worked together one after another.

One netizen provided ICML with a new idea: "another way to express this strategy is that all texts in submission should be generated by neural networks with parameters less than 100B or more than 1 trillion." "

Others pretended to be ICML judges and advertised ChatGPT: "as a judge of ICML and other meetings, I appreciate that authors use tools such as ChatGPT to polish articles." This will make their papers clearer and easier to read. (the post has been modified by ChatGPT) "

As for the rule, Professor Erik Brynjolfsson of MIT simply summed up: "this is a battle that must be lost." "

In addition to playing around, some people have seriously expressed their thoughts on ICML regulations.

"it's a strange rule, especially for an artificial intelligence conference," said Thomas Dietterich, former chairman of AAAI. We should welcome all articles that contribute to science and technology, regardless of whether the author has accepted the assistance of AI or not. "

There are also suggestions for ICML to improve:

Using large language models for revision is good for improving spelling and grammar. Grammarly,Notion,Google Docs and even Word use large language models.

It seems that the solution is to explain the reasons and uses of large language models, not to ban them.

It turns out I'm not the only one who uses Grammarly to check spelling and grammar (doge).

Of course, some netizens expressed their understanding of this, believing that the move was to protect the authority of the judges.

Netizen Anurag Ghosh commented: "I think ICML's requirement is to prevent papers that seem to be correct from being published." For example, five fabricated / AI-generated papers published in the field of machine learning. This will expose the shortcomings of peer review. "

Others think, "large language models are just tools, and what if they can produce higher-quality papers?" The main contribution of these studies comes not from large language models, but from those researchers. Should we ban researchers from receiving any form of help, such as Google search, or talk about it with people who have no interests? "

As the popularity of big language models has continued unabated in recent years, the ban proposed by ICML has sparked heated discussion among scholars and netizens, which is expected to continue for some time.

However, at present, there seems to be a question that has not been discussed. How to judge whether a fragment of an article is generated by a large language model? How to verify it, does it depend on the weight check? After all, if it is really an article generated by a machine model, no one will deliberately label "this article is automatically generated by a large language model," right?

What's more, research papers, such as articles with clear logic, clear structure and highly stylized language style, are simply the best place for large language models to play. Even if the original text is a little blunt, if it is used as an auxiliary tool, I am afraid it is difficult to identify, so it is difficult to have clear standards to implement this ban.

From this point of view, I am afraid that the burden on the reviewers of ICML this time will be a lot heavier.

Reference:

Https://icml.cc/Conferences/2023/CallForPapers

Https://twitter.com/ylecun/status/1610367976016064513

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era), edited by Xin Peng David

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.