How many of you have seen this honest ICLR submission? [D]

Explain Like I'm Normal

A researcher submitted a paper to ICLR that included a candid comment admitting they manually tweaked results to look better, likely forgetting to delete the note before submission. This has sparked a massive debate on Reddit about the 'hidden' data manipulation prevalent in academic AI benchmarks. It highlights the growing pressure on researchers to produce state-of-the-art results even when the delta is marginal.