[How to get an account / participate in contests?](http://atcoder.jp/posts/2) [AtCoder's Contest Format](http://atcoder.jp/posts/1) - [AtCoder's Testcases](https://www.dropbox.com/sh/nx3tnilzqz7df8a/AAAYlTq2tiEHl5hsESw6-yfLa?dl=0) - [AtCoder's Rating System](https://www.dropbox.com/sh/zpgcogxmmu84rr8/AADcw6o7M9tJFDgtpqEQQ46Ua?dl=0) AtCoder Race Ranking: [2023 Algo](https://img.atcoder.jp/file/gp30/race2023.html), [2024 Heuristic](https://img.atcoder.jp/file/gp30/race2024H.html)

Currently, in ABC, some users use generative AI to solve problems, which has slightly diminished the competitiveness and reduced the reliability of ratings. It is very difficult to completely separate generative AI from the current programming environment, and it should not be entirely banned. However, merely using generative AI to solve problems is neither valuable as a competition nor as proof of IT engineering skills, and AtCoder believes it should be prohibited. ### Rules For detailed rules, please check the following page: [AtCoder Rules against Generative AI ver20240607](https://info.atcoder.jp/entry/llm-abc-rules-en) To summarize the rules: - It is prohibited to directly input the text or images of problem statements provided by AtCoder into generative AI or other programs. - It is allowed to input into generative AI the information that a human has entered after reading the problem statement, or information that has been mechanically converted from it, and to use the output. - It is allowed to use generative AI for specific purposes such as translation, or to copy and use constants and proper nouns. These rules apply only to ABC. ### Guidelines for Rule Creation and Future Rule Changes The following two points were emphasized in the creation of these rules: - Minimize disadvantages for users who work on problems in the same way as before the advent of generative AI. - Ensure that participants can easily determine whether they are violating the rules. The point in these rules that "all information entered by humans is allowed" does not fully align with the first guideline. However, based on the second guideline, we had to either completely ban the use of specific software or set such rules, and we chose the latter. AtCoder hosts three algorithmic contests: ABC, ARC, and AGC. These three contests select problems based on their typicality rather than difficulty. ABC focuses on educational value and only features highly typical problems. Consequently, generative AI, which tends to be strong with existing problems, can solve about half of the problems. Using generative AI to solve ABC problems leads to the collapse of competitiveness, a decline in the overall contest experience for users, and a decrease in educational value. Simply giving the problem to generative AI and submitting the output code has no value in the competition or as an IT engineer. On the other hand, many people already use generative AI in today's programming environment, and it is expected that programming using generative AI will gradually become more common. AtCoder aims to deliver the competition of designing algorithms and the joy of coming up with them. This time, we rebalanced the competition by restricting the methods of providing information to generative AI. Currently, we do not consider restricting the use of generative AI for ARC/AGC. These contests are designed to require more flexible thinking than ABC. Consequently, past ARC/AGC problems are rarely solved by generative AI. Furthermore, from now on, problems that generative AI can solve on its own will not be selected for ARC/AGC. Therefore, even if there are situations where using generative AI is advantageous in ARC/AGC, we judge that it does not reach a level that undermines competitiveness. From another perspective, we also believe that competing in problem-solving abilities, including the use of generative AI, has meaning. For example, when debugging your program, it is common to write a slow exhaustive search solution and compare outputs. In this case, if you use generative AI to write the exhaustive search, it is just a smart use of AI without depending on it to solve the problem. Of course, if the exhaustive search solution can directly solve the problem, the situation is different, but such problems are not presented in ARC/AGC, so there is no impact. (Conversely, it can be said that such problems are presented in ABC, which is why the regulations are in place.) For these reasons, we decided not to restrict the use of generative AI in ARC/AGC. Regarding the actual possibility of detecting users who cheat in ABC with these rules, honestly, it is often impossible. In the first place, cooperation among multiple people online is also impossible to detect. We believe that competitive programming relies on the participants' conscience. We ask for your cooperation to protect the culture of competitive programming. If many users are clearly detected violating the rules, we are considering strict measures, including banning access from specific regions or countries to certain contests and taking legal action. Finally, depending on the future evolution of generative AI and the trends of many users, we may add further restrictions, such as banning the use of specific services. We intend to keep the rules easily accessible, so please check them regularly.