Subbarao Kambhampati proposed a conference affirmative action track.
Both #CVPR2026 awards went to industry-supported papers. Yet… it’s an academic conference, and therefore it’s odd one can flex “compute muscles” to win an award.
The meaning of “student” award is far gone. We need a “best academic award”
Brain over brawl?
Some users see industry papers dominating CVPR awards as a catalyst for universities to invest more in compute, while most criticize the wins as non-reproducible demos that undermine academic conferences and favor big labs.
@taiyasaki AAAAIR (Affirmative Action for Academic #AI Research) track at conferences...😋
@taiyasaki Agreed! That is exactly one of the reasons why we wrote the paper "Brains Over Brawn: Small AI Labs in the Age of Datacenter-Scale Compute" a few years ago... https://doi.org/10.1007/978-3-031-66705-3_2
@taiyasaki No idea how can a paper that is not reproducible win a best paper award, this is hillarious
@taiyasaki CVPR started to feel like the Crossfire map from Half-Life. Gates are closing and those who are inside don't honestly care what happens outside. Even if they are from the same team.
In light of the complaints that #CVPR2026 best paper awards went to industry-led ones, here again is a plea for AAAAIR (Affirmative Action for Academic #AI Research) track at conferences...😋
@taiyasaki Wasn’t the student award based on work using only one volta gpu?
@taiyasaki Agreed. It's not good for academic conference
@taiyasaki @aykuterdemml Something has to change.... But what ...
@taiyasaki If true, I see it as catalyst for universities to spend more on data centers/compute, so students have access to what they need. Stanford has $47B sitting around 😜
@Michiels_Nick @taiyasaki This would become more common in coming years. Scaling laws give an obvious advantage to giants and so they will dominate during early years because of their hands on infinite levels of compute. This would slow down once they start hitting bottlenecks
@taiyasaki IMHO, at the least, the best paper awards should be given to works with open implementations.
@taiyasaki Adding a 'best academic' category just creates a consolation prize. Papers needing 256 A100s to reproduce aren't science. They're demos.
Subbarao Kambhampati proposed a conference affirmative action track.
Both #CVPR2026 awards went to industry-supported papers. Yet… it’s an academic conference, and therefore it’s odd one can flex “compute muscles” to win an award.
The meaning of “student” award is far gone. We need a “best academic award”
Brain over brawl?
