Returning from EMNLP in Suzhou, Richard Diehl Martinez joins Pierce Freeman to explain the unusually important role of conferences in machine-learning research. NeurIPS, ICML and ICLR overlap in scope, while ACL and EMNLP serve language-focused work. Accepted proceedings papers are scholarly publications, not merely informal presentations.
OpenReview makes parts of peer review and discussion visible, but each venue determines its own access settings and decision process. The hosts discuss Andrew McCallum's contribution to the platform and ask whether transparency can compensate for overworked reviewers. They worry about a cycle of AI-written submissions receiving AI-assisted reviews, without establishing its prevalence.
Preprints provide a faster route to sharing ideas and establishing a dated public record. They do not confer peer-review approval, although the same manuscript may later appear in a reviewed venue. Interactive publishing and research blogs offer other ways to communicate methods that do not fit comfortably into a static PDF.
The cost debate separates access to research from attending and presenting at an event. Registration, travel and accommodation can create significant barriers, but the major proceedings discussed are freely readable. Corporate sponsorship brings recruiting opportunities and shapes the experience, with the hosts disagreeing about how much signal conference attendance supplies.
Their strongest shared defense of conferences is the chance to meet collaborators across institutions and countries. Publication counts, online visibility and portfolio code each offer incomplete evidence of someone's contribution. The episode ends by calling for better review incentives without discarding the value of research communities.
Watch on YouTube
