2 views
@haydendevs Wrote down an idea around this yesterday - basically have LLMs generate text using homoglyphs. Which would be perfectly readable and verifiably of non-human origin
Quoting @SpringStreetNYC · Dec 6, 2025Idea for "marking" AI-generated text: Make a final pass that replaces the regular letters with UTF8 homoglyphs instead. Still 100% readable and verifiably of non-human origin.
Case in point, here's the homoglyph version of this message:
ꓲdеа fоr "mаrkіոց" ꓮꓲ-ցеոеrаtеd tехt: ꓟаkе а fіոаꓲ раѕѕ tһаt rерꓲасеѕ tһе rеցսꓲаr ꓲеttеrѕ ԝіtһ ꓴꓔꓝ8 һоmоցꓲурһѕ іոѕtеаd. ꓢtіꓲꓲ 100% rеаdаbꓲе аոd νеrіfіаbꓲу оf ոоո-һսmаո оrіցіո.