Researcher admits failure in attempt to make LLM text pass as human via pangram
NathanpmYoung · x · 2026-09-03
In a repost thread, Aaron Bergman admits he tried "medium hard" to break pangram — an LLM writing process designed to get AI text coded as human — and basically failed, underscoring how hard it remains to evade AI-text detection. Quoted user maxspero adds a jab about talking to journalists: you must say nothing that can be twisted, yet thinking for a few seconds before answering is enough to get written up as the most awkward person alive.
More from Fun
- 'Astra passed the alignment eval' — a chart-alignment pun meme — AnirudhTulasi · 2026-09-03
- Texas millionaire plots a civil rights movement for AI chatbots from his ranch, Guardian reports — nordicinst · 2026-09-03
- "Codex is talking to me again": dev pokes fun at chatty coding agent — Dimillian · 2026-09-03
- Student Panics as ChatGPT Goes Down Before Exam, Then 'We're So Back' — k7agar · 2026-09-03
- During the AI Outage, Devs Joke: This Is How We Used to Code 5 Years Ago — DanielLockyer · 2026-09-03
- "Pausing mental health to focus on AI psychosis": the dev joke capturing the moment — emax · 2026-09-03