Saved in:
| Main Author: | |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.21653 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911703406477312 |
|---|---|
| author | Smirnov, Alexander |
| author_facet | Smirnov, Alexander |
| contents | AI text detectors amplify a pretrained typicality axis; they do not construct an AI-vs-human boundary. On raw encoders before any task supervision, projecting onto centroid(AI)-centroid(HC3) achieves NYT-vs-HC3 AUROC 0.806/0.944/0.834 across three architectures (86-106% of the fine-tuned discrimination ceiling: on RoBERTa-base, raw projection exceeds fine-tuning); on RoBERTa-base, full fine-tuning reduces discrimination below raw on both fluent-formal populations tested. The same axis inverts on non-native ESL writing (AUROC 0.06-0.20) -- a falsifiable prediction unique to the typicality reading. A 24-example frozen probe matches full fine-tuning (0.900 vs 0.895). A closed-form Jacobian predictor parameterises axis-manipulating interventions with R^2 = 1.000 universal, lifts ELECTRA-CE deployment TPR from 0.000 to 0.904 at FPR = 1%, and transfers to three independently-trained third-party RoBERTa detectors at 16/16 oracle-equivalence (57% NYT-FPR reduction on the OpenAI detector). Scope: encoder family; mechanism magnitude HC3-anchored; population-level shared axis with per-text mechanisms varying across architectures. Three operationally distinct probes -- text-surface caps_rate residualisation, geometric signed-epsilon ablation, closed-form text-pair predictor -- agree at cos 0.74/0.81/1.00 across three architectures, confirming observer-invariance. Under matched-TPR-0.90 evaluation, the published intervention zoo (CC, dealign-f2c) is calibration-equivalent across 27 cells (|Delta AUROC| <= 0.0081), and >= 97% of the LoRA->full-FT bias gap on ELECTRA is calibration shift, not learned representation -- the central claim's prediction confirmed. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_21653 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction Smirnov, Alexander Machine Learning Artificial Intelligence Computation and Language AI text detectors amplify a pretrained typicality axis; they do not construct an AI-vs-human boundary. On raw encoders before any task supervision, projecting onto centroid(AI)-centroid(HC3) achieves NYT-vs-HC3 AUROC 0.806/0.944/0.834 across three architectures (86-106% of the fine-tuned discrimination ceiling: on RoBERTa-base, raw projection exceeds fine-tuning); on RoBERTa-base, full fine-tuning reduces discrimination below raw on both fluent-formal populations tested. The same axis inverts on non-native ESL writing (AUROC 0.06-0.20) -- a falsifiable prediction unique to the typicality reading. A 24-example frozen probe matches full fine-tuning (0.900 vs 0.895). A closed-form Jacobian predictor parameterises axis-manipulating interventions with R^2 = 1.000 universal, lifts ELECTRA-CE deployment TPR from 0.000 to 0.904 at FPR = 1%, and transfers to three independently-trained third-party RoBERTa detectors at 16/16 oracle-equivalence (57% NYT-FPR reduction on the OpenAI detector). Scope: encoder family; mechanism magnitude HC3-anchored; population-level shared axis with per-text mechanisms varying across architectures. Three operationally distinct probes -- text-surface caps_rate residualisation, geometric signed-epsilon ablation, closed-form text-pair predictor -- agree at cos 0.74/0.81/1.00 across three architectures, confirming observer-invariance. Under matched-TPR-0.90 evaluation, the published intervention zoo (CC, dealign-f2c) is calibration-equivalent across 27 cells (|Delta AUROC| <= 0.0081), and >= 97% of the LoRA->full-FT bias gap on ELECTRA is calibration shift, not learned representation -- the central claim's prediction confirmed. |
| title | Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2605.21653 |