NeurIPS 2026 Main Track — Theory papers score tracking post Rebuttal [D]

The recent post on Reddit’s NeurIPS forum, sparking a conversation about theory paper scores following the rebuttal period, highlights a recurring tension within the AI research community: the evaluation of theoretical contributions. The request for score distributions, particularly for theory papers, isn't simply about data aggregation; it reflects a deeper concern about the perceived undervaluation of theoretical work compared to empirical results. This aligns with observations made in our own publication, like the piece exploring the challenges of managing AI Slop Is Costing You Hours. Here's How To Stop Sending It, where the focus on practical application can inadvertently overshadow the foundational work that enables those applications. The user’s self-reported scores of 4/4/4, with a confidence of 3/3/3, while seemingly positive, are framed within a context of generally lower scores this year, raising questions about the overall rigor and consistency of the review process.

The discussion resonates with the ongoing debate around the balance between theoretical advancements and demonstrable impact. Often, theory papers – those focusing on establishing new mathematical frameworks, proving fundamental limits, or developing novel algorithmic principles – are judged on their immediate applicability, which can be difficult to assess. This can lead to a bias towards papers showcasing tangible results, even if those results are built upon less robust theoretical foundations. The original poster’s observation that theory papers "often seem to get somewhat lower scores" is a sentiment shared by many within the field. It's a reminder that the OpenReview system, while aiming for transparency A question on ICLR and NeurIPS deadlines, and OpenReview [D, isn’t a perfect solution for nuanced evaluation, particularly when dealing with abstract concepts. Even groundbreaking theoretical work, like the recent explorations in “Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation”, Gladstone et al. 2026 [R, may require a longer timeframe to fully realize its implications.

The lower scores across disciplines this year, as noted by the poster, further complicate the picture. It suggests a broader shift in reviewer expectations or a potential increase in the volume of submissions, making the evaluation process even more challenging. The call for anonymous or approximate score sharing underscores the sensitivity surrounding these discussions and the desire for honest feedback without fear of retribution. This emphasis on transparency, while commendable, also highlights the inherent subjectivity in peer review and the potential for bias, even among well-intentioned reviewers. It’s a space where constructive criticism and a deeper understanding of the nuances of theoretical research are vital, not just for individual researchers but for the overall health of the AI community.

Ultimately, this conversation serves as a valuable reminder that the pursuit of AI progress isn't solely about achieving immediate, demonstrable gains. A robust theoretical foundation is crucial for long-term innovation and the development of truly intelligent systems. Moving forward, it will be interesting to see if the community develops more refined mechanisms for evaluating theoretical contributions, perhaps incorporating more specialized reviewers or utilizing metrics that better capture the significance of abstract concepts. Will we witness a renewed appreciation for the foundational work that underpins the increasingly complex AI landscape, or will the pressure for immediate impact continue to overshadow the importance of rigorous theoretical exploration?