# Summary of Author Reviewer Rebuttal Discussions

Dear Acs,

We greatly thank you for your time and efforts in carefully handling our paper. As the rebuttal-discussion phase is coming to an end, we would like to quickly summarize the rebuttal discussions and reiewer's feedback as follows.

Reviewer hBT7 (***raised score to 7***) mentioned that "*I find the research very interesting***" and believes that our work "*is *an important contribution** that needs to be **heard by a larger audience**, especially in a conference*". The reviewer has also *appreciated the organization*** of the paper "*The visualization (Figure 1) placed early on the paper also helps the reader to further contextualize the problem.*"

Reviewer 8upM (***raised score to 6***) appreciates that *"paper investigates an *important problem**"* and "*paper is well-written*/presented". The major concern of the reviewer was regarding the base language models used and to address the same, we performed detailed experiments with various SOTA LLMs in our rebuttal and show our results holds true. This addressed all the reviewer's concerns and the reviewer raised the score appreciating our efforts.

Reviewer NSUG (***score 5, no response to rebuttal***) believes that the paper *"addresses an important and timely topic in the field of NLP"* and "*the author(s) discovers the **hidden possibility** and provides a theoretical analysis*". The reviewer had only minor issues and doubts regarding the analysis and motivations of using Large deviation theory. We addressed all the concerns by providing mathematical and intuitive explainations of our analysis. We also addressed the reviewer's doubts with additional experiments in the rebuttal.

Reviewer LTNg (***score 4, no response to rebuttal***) mentioned that "*results presented in this work are general and *interesting*" and "*They *address the important drawback* from the previous work*". The reviewer's primary concern revolved around the experimental setup i.e of using GPT-2 LLM and simple detectors. We addressed all the concerns of the reviewer by performing detailed experiments with various SOTA LLMs and SOTA detectors in our rebuttal and show that our results holds true.

Reviewer RcT9 (***score 3, no response to rebuttal**): We appreciate the reviewer's feedback and efforts in reviewing our paper. But we are worried that Reviewer RcT9 does ***not provide a fair assessment*** of our work. Most of the comments are quoting sentences/part of sentences from our draft and are not technical questions of our work or analysis. We had shared our concerns in detailed manner in earlier request to ACs [[link to request]](https://openreview.net/forum?id=lMdl4XsvmQ&noteId=ugSSpbNVYW). We had also shared the dataset with comments as the reviewers had asked. However, the reviewer hasn't engaged and responded with our responses.

>Note : Most of the reviewers have acknowledged that our work provides an optimistic perspective on detection by highlighting the various hidden possibilities which is very critical.

Thank you once again for your time, efforts, and dedication to the review process. Your feedback and active involvement in the review process is much appreciated.

Regards,

Authors

Reviewer KNFz appreciates that "this paper clearly defines an important problem which has been a long-understudied problem in RL".

Reviewer DaQT mentioned that "the proposed approach is technically interesting" and "convergence analysis seems correct and could be of wider interest".

Reviewer 65in believes that "The ideas in the paper are novel" and "mathematical formulation and the derivation presented are unique and interesting".

Reviewer fDEr mentioned that "The paper tries to address an important limitation that arises because of the approximation required in the inner optimization process of the implicit gradient-based bilevel optimization"