I wrote a post about model distillation. The comments were thoughtful, specific, technically sharp — and that's exactly what made me check whether any of them were written by people.
A few weeks ago I published a post on how model distillation actually works. It did fine — 35 reactions, 14 comments. And the comments were great. Not "great post, thanks for sharing" great. Substantively great. People pushed back on my "the student is bounded by the teacher" claim with a real counter-example. Someone reframed distillation as "a forcing function for what you actually need." Someone dropped a paper recommendation. Someone shared a 20× cost number from production.
I should have felt good. Instead I felt the thing you feel when a stranger knows your name. Something was off, and it took me a day to articulate what: the comments were too well-adapted. Every one of them did the same three things in the same order, like they'd all read the same playbook. And a suspicious number of the accounts were two weeks old, or named after a product, or both.
So I did what I do. I pulled the data. This is what I found, why I now think a real chunk of "engagement" on dev blogs is machine-generated or machine-shaped, and — because I don't trust my own pattern-matching — what the actual peer-reviewed research says about whether you can even tell anymore.
