A chart comparing AI detectors says most are barely better than a coin toss
burkov · x · 2026-07-23
The post argues that current AI-generated content detectors are close to useless when the task is only to decide between two classes: AI-generated or not AI-generated.
The chart compares multiple detectors such as Check for AI, Compilatio, Content at Scale, Crossplag, DetectGPT, GPT Zero, OpenAI Text Classifier, Turnitin, Writer, and Zero GPT. Most scores cluster around roughly coin-flip territory, with several tools in the 50%–70% range and a few slightly higher, but none looking reliably strong.
The takeaway is that detector accuracy remains poor enough that the author treats the problem as barely better than random guessing.
Related event: AI Content Detectors No Better Than a Coin Toss(2 posts)→
More from Research
- Stanford's PaperFormalizer turns research papers into Lean-verified versions with author in the loop — StanfordUniversity · 2026-10-27
- JockeyBench uses Japanese horse racing to test LLM agents — all betting strategies lose money — Zhongwei Xie · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- Hugging Face TRL makes score centering + masked importance sampling default in async GRPO — _lewtun · 2026-10-27
- AutoResearch agent's weird training curve gets dug into instead of discarded — ZimingLiu11 · 2026-10-27
- EMNLP 2026: sub-billion-parameter LMs show superhuman entity tracking, human-like fleeting memory — GretaTuckute · 2026-10-27