What's this about?
Mollick: for a brief period many people were better off having Claude do a lot of their writing, because it is a pretty good writer and AI detectors were bad. That window has closed — ClaudeSpeak is now clichéd, suspect and annoying, and Pangram is well known.
What is "ClaudeSpeak", and what is Pangram?
ClaudeSpeak is the recognisable register: the measured hedging, the tricolons, the "it's not just X, it's Y" construction, the relentless evenhandedness. It was invisible when rare and is now a tell, in the same way stock photography became visible once everyone used the same library. Pangram is an AI-text detector with a considerably better reputation than the early ones that produced false accusations against students — so the two halves of the arbitrage closed simultaneously: the output became recognisable to humans *and* to detectors.
What's the community saying?
- The framing as an *arbitrage* is the useful part: the gain came from a temporary asymmetry, not from the writing being good.
- It sits alongside his point from last week that using a weak model for human-facing content will read as disrespectful — the standard is moving in both directions at once.
- "First Golden Age" implies a second, which is doing quiet work in that sentence.