Slop is text that's grammatically correct but says almost nothing (vague, filler, repetitive). AI made slop more abundant, but it also exposed how much slop was already there in human-written text.

Slop can be made conditional on context. Conditional slop is how sloppy a piece of text is given everything else around it.

ML papers are full of this. Given the rest of a paper, these are often conditional slop:

• The intro's first paragraph: "LLMs can now do this and that..."
• Theorems that are technically correct but obvious from their assumptions
• Experiments: if your model is good at MMLU and HLE, then it will crush GPQA too
• Conclusions restating the abstract in past tense

References

• Measuring AI "Slop" In Text