Go to z.umn.edu/QMRAbot and create an account by signing in with the checkbox
"Create Account" checked.
Go to the Account page. You can upload to “Local Database” the files that contain Basic Knowledge of your research field, such as review
articles, your own papers, major references for your active projects, etc. It is not necessary for you to upload anything to the
“OpenAI Database”. Note: You will see a default file in both databases, which is a file recording useful information from your future
conversations with the QMBot.
Go to the Chat page to start a conversation.
Chat with the QMBot (instructions are given in the next section).
When you are done with a conversation, click “New Chat” to start a new conversation thread or “Log out”.
How to Chat with the QMBot:
You can ask the ChatBot to do a search on arXiv, e.g.: Get me recent 10 papers about superconducting materials from arXiv.
Once the search is done, you will see the list of retrieved papers in the chat box and also in the table of “Retrieved Papers” on the
right side of the chat page.
There are three columns in the table, each showing the title of the paper with a clickable link to their arXiv page, the
status/action of associating the paper to the current conversation, and a numeric value that reflects the relevance of this
paper as compared to the Knowledge Base (files in Local Database). A lower value reflects a shorter linguistic distance thus
a higher relevance.
You can click on the “link” in the chat box or title in the “Retrieved Papers” to access a paper on the arXiv website.
If you are interested in one or more papers and want to chat more about their detailed contents, click on the “Associate”
button in the second column of the table. Once associated, the orange “Associate” button will turn to a red “Delete” button
to allow you to remove a paper from the current conversation.
Please note, after association if the messages do not have an enhancement notification, it did not retrieve the data from the file contents, and the ai only used the abstract. This can be
fixed by adding "use the vector store" to the end of your prompt.
More in-depth questions you can ask the QMBot include: Look into the full text of the paper titled “xxx” and give me a summary; Go
through the papers titled “xxx”, “yyy”, “zzz”, … and make a table summarizing all the transition temperatures of antiferromagnets
mentioned.
During the conversation, if you find some pieces of information useful to be added to your knowledge, you can say
“Remember this”. The QMBot will record a summary of the conversation in a user preference document.
How the QMBot is Personalized:
On the Account page, Local Database files contain central knowledge to the user, which will be used to calculate relevance scores
(linguistic distance) for search results returned from arXiv.
Still on the account page, the “OpenAI Database” contains files associated with the current conversation. These files are what you can
ask detailed questions about.
You can manually add papers to the OpenAI Database, so you can include your previously familiar papers in the current conversation,
together with the papers newly fetched from arXiv.
The user preference document under both the Local Database and OpenAI Database registers your previous conversations with the QMBot,
in which the contents will also be used to calculate relevance scores.
The OpenAI database (except for the user preference file) will be cleared after logging out.