The seminar schedule will be added here as it is planned. Any participants interested in leading a week's discussion on a topic they consider especially important should email the organizers.
25/09 · The organizers
Community debrief: How has AI been useful, harmful, or surprising in your mathematical work, daily activities, or life?
To open the series, several members of the community volunteered to give 3–5 minute short stories about a personal experience with, emotional reaction to, or prediction for AI. The goal was to map out our collective perspectives, interests, and concerns to guide future sessions. Topics of particular interest emphasized by the community include: the future of mathematical publication, issues of AI plagiarism, bootcamps on the use of AI for mathematical research and formal verification, and misalignment and breakout incidents by AI.
02/10 · Cass Halleck-Dubé (MPIM)
Community polls on AI use and policy.
In this meeting, we polled the attendees on a range of issues relevant to community use of AI. The full results of the poll can be found here.
09/10 · David Prinz (London Initiative for Safe AI) and Andrin Hirschi (Bonn)
AI Usage Classification
Two versions of an AI usage classification will be presented and then discussed. In David Prinz's proposal, authors declare how much AI contributed to each aspect of their work, from language editing to theorem proving, on a scale from 0 to 4. In Andrin Hirschi's proposal, the amount of human understanding in a paper is classified in one number.
16/10 · Multiple volunteers
Digesting the OpenAI preprints: Legibility, Correctness, and Insight
Following the sudden release on October 6th by OpenAI of claimed proofs of nearly 400 mathematical results, we invite multiple volunteers to investigate one of the preprints in an area relevant to their research, and to present their findings with regard to the following questions.
- Is the result true? (to the best of their judgment)
-
Are the essential insights of the proof intelligible?
- Is the document written in a way consistent with community standards?
- Were existing papers and prior literature related to the problem appropriately cited?
- Is OpenAI's treatment of these results consistent with the guidelines of the IAS advisory group for the release of AI-generated mathematical results?
After the volunteers present their experience and findings, we will proceed with a community discussion. We will target both 1) strategies to digest these papers and 2) recommendations for OpenAI or the advisory board on how to improve the process.
If the result investigated has an associated Lean proof, volunteers with experience in Lean are encouraged to investigate (possibly with AI assistance) whether the Lean proof reflects the same argument as the informal proof.