Has AI Tells

Commenters debated tools and heuristics for spotting “AI tells” in writing, noting existing projects like Simon Willison’s cliche highlighter, Adam Dunkels’s is-it-slop, and research such as SlopShape (corysama) that claims to fingerprint different models. Isamu urged shifting focus away from origin detection toward rewarding clearer, less templated writing, and arjie pointed out that simple rule-based detectors can be reproduced from the article’s description. Others, like zanderwohl and TimPC, complained less about origin than about homogenized style and overused patterns; some readers even preferred minor human typos as a signal of authenticity (zanderwohl).

Opinion splits over whether origin matters versus quality and the broader social effects. Zambyte, twosdai, and TimPC accept AI for practical writing (documentation, technical posts) if it’s accurate, while programjames and AlienRobot demand provenance and worry about hallucinations and reduced “comprehension bitrate.” Bigstrat2003 and acedTrex assert intrinsic human value in output, whereas socializer argues for cordoning off AI content to prevent automated crowding out human expression. Others raised practical concerns - detectors being gamed (zanderwohl), mediocre model output by design (palmotea), labor-theft worries (GolfPopper), and the need for experimental studies rather than surveys (pessimizer).