Funny how the tell keeps relocating. The delves and the em-dashes faded, so now it's the substrate-and-nodes voice, and in a year that one will read as dated too. Spotting "this is AI" stays a moving target, because the second a tic becomes easy to name, it gets tuned out and a new one grows in.
lol in the middle of this when I got to that Claude-ified paragraph I was like "oh no, I hope Benn is ok because what is going on with this AI-ass paragraph" and then you immediately addressed it
Loved the “quilt of empty nonsense” paragraph as a demo of the problem. It got me thinking: for someone who doesn’t have the domain knowledge or experience, how do you practically tell the difference between (a) genuinely deep, hard‑to‑parse insight and (b) this kind of statistically plausible, pseudo‑profound word salad?
Are there any heuristics you use yourself when reading frontier‑model output (or even human writing) to decide “this is worth untangling” vs “this is just noise dressed up as intelligence”?
Funny how the tell keeps relocating. The delves and the em-dashes faded, so now it's the substrate-and-nodes voice, and in a year that one will read as dated too. Spotting "this is AI" stays a moving target, because the second a tic becomes easy to name, it gets tuned out and a new one grows in.
lol in the middle of this when I got to that Claude-ified paragraph I was like "oh no, I hope Benn is ok because what is going on with this AI-ass paragraph" and then you immediately addressed it
Also if I could go the rest of my life never hearing the words load-bearing, substrate, or clobbered again, it would be too soon... !
the orthogonal axis is the load-bearing one... well played.
Loved the “quilt of empty nonsense” paragraph as a demo of the problem. It got me thinking: for someone who doesn’t have the domain knowledge or experience, how do you practically tell the difference between (a) genuinely deep, hard‑to‑parse insight and (b) this kind of statistically plausible, pseudo‑profound word salad?
Are there any heuristics you use yourself when reading frontier‑model output (or even human writing) to decide “this is worth untangling” vs “this is just noise dressed up as intelligence”?
Love your writing Benn