AI text detection will never be fully solved, but current tools are already very accurate
CamachoCollados · x · 2026-11-07
An NLP researcher posted a thread clarifying common confusion about the state of AI-generated text detection, arguing all of the following can be true simultaneously:
- AI generation detection is not fully solved and never will be;
- Current detection tools can nonetheless achieve very high accuracy;
- They are quite useful, especially amid the massive influx of AI slop;
- But they cannot be used as sole evidence, since they do make mistakes and are usually opaque;
- AI-generated text != bad text.
He adds that most AI-generated text today is indeed bad (usually correlated with the author's lack of care) and calls for better solutions to the problem as a whole.
More from Models
- Decision models fix a pricing inefficiency: prefill workloads subsidize decode, argues sh_reya — hugobowne · 2026-11-07
- shadcn reports weird eval results: Sonnet and Gemini pass the most tasks — shadcn · 2026-11-07
- User generates full rocket launch video with GPT-6 Pro in a plain ChatGPT chat — VraserX · 2026-11-07
- No new model has dethroned Opus 5.5 as big releases stall — haider1 · 2026-11-07
- Nine AI agents tackle national lab open questions; none pass expert review — The Decoder · 2026-11-07
- OpenAI mass bans hit account resellers; DeepSeek traffic surges as users appeal wrongful bans — huangyun_122 · 2026-11-07