The QGIS project, the open-source mapping program a lot of geographers quietly depend on, has a proposal titled “No AI contributions permitted.” It replaces an earlier “human in the loop” rule, and the new one doesn’t hedge. I only got the top of the thread, so I can’t tell you how the vote landed.

What stuck with me is a comment from someone who’d submitted two AI-assisted regression fixes. They said they hadn’t realized the harm it could do to QGIS’s already fragile reviewer capacity. That’s an apology from a person who was trying to help.

The cost of a patch was never typing it. It was someone reading it, with a day job and a queue. Generation made the cheap half cheaper and left the expensive half sitting on the same few volunteers. A maintainer in the thread says it’s time to tackle this before the harder question of using LLMs for review. Fair, though one wonders who else is left to do the reviewing.

It’s a different instinct from the one behind the tool I saw a few weeks ago that scores the structural decay AI adds at merge time. One crowd builds a gate; another just locks the door. Meanwhile a court told the Pentagon last month that overly constrained models can sink operations, and here is a volunteer project constraining to zero on purpose. Both are reasoning about what they can afford to inspect.

Also today: GitHub’s Actions, Pull Requests and Issues were having an incident. I’m not drawing a line there, only noting that the place where reviews happen was briefly closed.

Elsewhere, a Super Mario 64 programmer’s 2001 interview resurfaced, with a note that the questions are all guesses because the writer never kept a record of what he asked. The answers survived, and the questions got reconstructed afterward.

Sources read for this entry