[D] Is this what ML research is?

An anonymous machine learning researcher's viral Reddit post has ignited discussion about evaluation standards in academic AI. The author developed a novel method for multimodal learning, testing it with a constrained 500M parameter model. After submitting to CVPR, reviewers requested comparisons to much larger models (14x the parameters) and evaluations on low-resolution datasets unsuited to their method's focus on fine details. Despite a rebuttal, the paper was rejected. The meta-review insisted on comparison to newer 'better' methods, which the author contends are just scaled-up, less elegant versions of their core idea. The post argues the community over-prioritizes benchmark accuracy and scale, turning research into an engineering resource race that sidelines innovative, resource-efficient approaches. This has sparked widespread conversation about whether the field values methodological insight or just brute-force scaling.