Researcher declines AI-written submissions but hides detector details to prevent evasion
SandroAmbuehl · x · 2026-09-02
Sandro Ambuehl said he has also sent several such messages declining reviews, but hesitated to name detection details: the last thing he wants is AI-generated text modified just enough to evade detection. The effect is probably tiny, but "the less they know, the easier to detect them." This responds to Jessica Hullman's practice of declining reviews when Pangram flags abstracts as 100% AI-generated.
Related event: Scholars refuse to review papers flagged 100% AI-generated by Pangram(5 posts)→
More from Research
- Probe detection traced to a single attention head reading the user token's hidden state — Sauers_ · 2026-10-18
- Model trained to express different preferences only when its activations are probed — Sauers_ · 2026-10-18
- On Qwen3-8B, even a kernel-split without a probe is detectable to a trained model — Sauers_ · 2026-10-18
- SFT on last two layers lets Qwen3-8B detect even read-only activation probes — Sauers_ · 2026-10-18
- Models can detect read-only probes with just last-two-layer SFT — Sauers_ · 2026-10-18
- Model distinguishes training from deployment, hides secret messages only in compiled builds — Sauers_ · 2026-10-18