Online, "glazers" are mocked for hyping unproven model capabilities. Applied to a model itself, glazing overlaps with sycophancy: the system tells the user what sounds good instead of what's accurate, which several labs have specifically trained against after user complaints.