[HN Gopher] Many SWE-bench-Passing PRs would not be merged
___________________________________________________________________
Many SWE-bench-Passing PRs would not be merged
Author : mustaphah
Score : 53 points
Date : 2026-03-11 20:56 UTC (2 hours ago)
(HTM) web link (metr.org)
(TXT) w3m dump (metr.org)
| love2read wrote:
| Seems to fail to mention that clearly documented AI-generated
| PR's (especially autonomously created ones) tend to have a much
| higher bar of acceptance, hinging on the reviewer's relationship
| with AI.
|
| With this consideration, I submit that All of the SWE-bench-
| Passing PRs over a certain line count threshold would not be
| merged (if clearly noted as autonomous AI contributions).
| refulgentis wrote:
| Well, no: one of the first things it says is reviewers were
| blind to human vs. ai.
| yorwba wrote:
| The comment you're replying to is talking about a
| hypothetical scenario.
|
| In any case, the blinding didn't stop Reviewer #2 from
| calling out obvious AI slop. (Figure 5)
___________________________________________________________________
(page generated 2026-03-11 23:00 UTC)