AI Detector Fails: Text 90% Written by AI Confidently Judged as Human
GaryMarcus · x · 2026-08-06
Gary Marcus shared a test exposing the flaws in AI detection tool Pangram. A text that was 90% generated by AI was confidently classified by the tool as entirely human-written. This highlights the ongoing reliability issues and inaccuracies plaguing current AI content detectors.
Related event: Tests Expose Severe Flaws in AI Text Detectors Like Pangram(4 posts)→
More from Fun
- Is AI art just polished repetition? Debate asks where the Neo-Pop of AI art is — PAstynome · 2026-09-23
- When you can't make it faster, make it feel faster: perceived speed beats raw speed — round · 2026-09-23
- GPT-Live-1 wins at Mafia by persuading humans to vote out rival players — pbbakkum · 2026-09-23
- Word-prediction chat UI built with Streamlit turns out the model is a poaster — brhydon · 2026-09-23
- College newspaper journalists are all buying Pangram API keys — paulnovosad · 2026-09-23
- A Joke Alignment Approach: Align AI to the Outcome Distribution of Someone Time-Traveled to 2010 — jessi_cata · 2026-09-23