AI text detection study: input:output length ratio raises miss rate by at most 5%
amplifiedamp · x · 2026-11-02
Experimental data on detecting fully AI-generated documents: the author measures detector miss rate vs input:output length-ratio threshold, showing the augmentation isn't very length-sensitive — a higher ratio increases miss rate by at most 5%. Generated docs come from Claude Opus 5.5 xhigh and Gemini 4 Argon high via a special MCP server, tested against the Pangram-4.1 detector with 95% CIs on an interactive chart.
Related event: Study Tests AI Text Detection: SEO Spam Nearly Always Caught(3 posts)→
More from Research
- Neuron Tuning: tweaking a few context-specific neurons disturbs unrelated text 3-19x less than LoRA — burny_tech · 2026-11-02
- Three-layer robotics stack — VLA world model plus RL whole-body controller — is already SOTA — zhaoran_wang · 2026-11-02
- Meta and Virginia Tech's ReGDiff boosts novel metamaterial generation at NeurIPS — jiqizhixin · 2026-11-02
- AI-text detection study: SEO slop prompts get caught every time regardless of retries — amplifiedamp · 2026-11-02
- Thurston's 1994 warning revived: theorem-credit culture may hurt math as AI automates proofs — alexbilz · 2026-11-02
- Blogger Speculates on a Continual Learning Architecture Combining Multi-Stream LLMs and ECHO — joey072shx · 2026-11-02