Commentary on X
I see. That makes a ton of sense and yes the differences can be everything from visual to downright jarring. I don’t receive the level of email you likely do so I haven’t had to deal with the slop at the level. Even for long messages I don’t use AI when it’s personal in nature. Friends don’t mind my running sentences, etc but for professional settings I like to at least try and create grammatical, well paced prose. Admittedly even with AI I don’t always succeed. Your post made me think about my own use of AI is writing in a deeper way though. Always a good thing. 👍 Cheers. 🍺
2:17 PM · Jul 14, 2026These pattern detectors cannot tell you if the text is written by a human; they use a weak resemblance score masquerading as certainty. Humanizers bypass them, human writing gets flagged as a false positive, and they profit from suspicion infrastructure installed by lying that they detect AI.
10:36 AM · Jul 14, 2026@gneubig Why do you believe these AI detectors? You are much more educated on this. But I consider them all inherently unreliable and the marketing close to fraudulent. If there was a reliable signal, this would be great input for the next LLM training round.
6:22 AM · Jul 14, 2026@gneubig It’s kinda sad that we are reaching a point where even for emails, we have to check for slop.:(
7:06 AM · Jul 14, 2026View all
@gneubig Honestly yeah, most "personalized" emails feel more robotic than the templates they replaced.
10:45 AM · Jul 14, 2026It's so refreshing to get a human-written email nowadays.
6:12 AM · Jul 14, 2026@gneubig Just update a generated text here and there and engram will claim it's 100% human written.
4:06 PM · Jul 14, 2026