Moonlight provides a paper review page to help you understand newly published papers more quickly. The goal is to help researchers quickly understand the overall context before reading the paper by organizing the key content and contributions of the paper and helping them look at related research together.
I recently received an email from a researcher.
I read his paper review page and was impressed by the fact that he understood the context of the paper much better than the review service (Stanford Agentic Review) of a famous overseas university. It was really gratifying feedback for the team creating the product.
But for us, the advice that followed was even more impressive.
You write: stop struggling with complex Lie algebras and let AI work for you. I am sure it was meant as a light-hearted comment. Everything is a result of personal struggles and we do mathematics to satisfy our desire to know. We cannot delegate understanding to a machine.
(I understand that the expression “don’t wrestle with complex Lie algebras and leave it to AI” is a phrase put lightly. However, everything is the result of our own consideration and exploration, and we research because we want to know. Understanding cannot be delegated to machines.)
I re-read the email and saw that this recommendation was used on our review page.
Stop struggling with complex Lie algebras and let AI work for you. (Don't wrestle with complex Lie algebras; leave it to AI.)
The original intention was for AI to organize complex research and aid understanding. But when I read the email, it was a little different from the message we really wanted to convey.
We don't want to make Moonlight just an AI Assistant that answers questions. However, we do not want to create an AI Agent that performs research on our behalf. This is because I believe that AI should not replace the understanding researchers gain from directly encountering papers.
What we want to createAI Colleague researching togetherIt is.
What Moonlight focuses on is not how much work AI can do for you. It helps researchers think of better questions in the process of reading papers, grasp claims and evidence more clearly, and decide more quickly and confidently what to read and judge next. Furthermore, we dream that Moonlight will not only stop at the moment of reading a paper, but also accompany researchers throughout the entire process of organizing ideas, exploring related research, and deciding on the next research direction.
So I didn't just edit one phrase and call it a day.
First, I looked again at the suggested phrase creation prompt to see why this phrase was created. We also modified the blurb on the review page to be closer to the message Moonlight wanted to convey to researchers.
After reflecting the changes, I sent another email to the researcher. Thanks to the feedback, we revised the recommendation text and also conveyed our idea that Moonlight wants to create colleagues who help research, not AI that replaces research.
This incident made us think once again that the message we want to convey should be better included in the recommended text. Moonlight will not continue to be an AI that replaces researchers.Colleagues who make research deeper and more enjoyableI want to become.
From the standpoint of product development, feedback that tells us what attitude and language we should use to approach researchers is just as important as opinions on functionality. This email was also an opportunity that made me take another look at Moonlight.