E5-small LoRA AI detector accuracy test
The E5-small LoRA detector flagged exactly half of the human texts in our test as AI (50.0%), including 75.0% of human forum answers, while catching 61.1% of AI texts. Its AUROC of 0.582 means its scores barely separate AI writing from human writing.
Tested
Verdict: Fast and tiny, but it flags half of all human writing, so its verdicts carry almost no information.
E5-small LoRA detector test results
| Rank | 5 of 7 |
|---|---|
| Balanced accuracy | 55.6% |
| Catches AI text | 61.1% (308 of 504; 95% interval 56.8 to 65.3%) |
| Flags human text | 50.0% (63 of 126; 95% interval 41.4 to 58.6%) |
| Catches humanized AI text | 47.6% (60 of 126) |
| AUROC | 0.582 |
| Speed on an Apple M4 Pro | 0.03 seconds per 1,000 words |
| Type | Trained classifier |
| Threshold | 0.5 (default probability cut-off) |
| Licence | MIT |
| Model | MayZhou/e5-small-lora-ai-generated-detector |
Numbers behind this chart
| Item | Rate | Texts | 95% interval |
|---|---|---|---|
| Claude Fable 5.1 | 49.2% | 62 of 126 | 40.6 to 57.8% |
| Claude Haiku 5.5 | 66.7% | 84 of 126 | 58.1 to 74.3% |
| Claude Opus 5.5 | 64.3% | 81 of 126 | 55.6 to 72.1% |
| Claude Sonnet 5.5 | 64.3% | 81 of 126 | 55.6 to 72.1% |
| Humanized Claude Opus text | 47.6% | 60 of 126 | 39.1 to 56.3% |
Numbers behind this chart
| Item | Rate | Texts | 95% interval |
|---|---|---|---|
| Academic abstracts | 36.0% | 9 of 25 | 20.2 to 55.5% |
| Q&A answers | 75.0% | 21 of 28 | 56.6 to 87.3% |
| Encyclopedia articles | 52.0% | 13 of 25 | 33.5 to 70.0% |
| Fiction | 12.5% | 3 of 24 | 4.3 to 31.0% |
| News reports | 70.8% | 17 of 24 | 50.8 to 85.1% |
Numbers behind this chart
| Item | Rate | Texts | 95% interval |
|---|---|---|---|
| under 200 words | 30.9% | 30 of 97 | 22.6 to 40.7% |
| 200 to 299 words | 60.2% | 150 of 249 | 54.1 to 66.1% |
| 300 words or more | 81.0% | 128 of 158 | 74.2 to 86.4% |
What it is
This is the smallest model in our test, an E5-small text encoder fine-tuned with LoRA on the RAID dataset of human and AI writing. We used the standard probability threshold of 0.5.
Results
It flagged 50.0% of human texts: 75.0% of forum answers, 70.8% of news reports and 52.0% of encyclopedia articles. Only fiction got through mostly unflagged, at 12.5%.
It caught 61.1% of AI texts, and like Fakespot it did best on encyclopedia and news writing, at 92.0% and 94.8%. Those are the same genres where it flags human writing most, which points to a detector reacting to genre style. Its AUROC of 0.582 confirms that its scores barely separate AI from human writing.
Longer human texts were flagged more often: 64.3% of human texts of 300 words or more, against 39.5% under 200 words. A humanizer rewrite lowered its catch rate on Claude Opus 5.5 texts from 64.3% to 47.6%.
Human writing E5-small LoRA detector flagged as AI
It is extra payments that you apply to the principal. Here is an example. Let's say that you have a mortgage with a monthly payment of $1000 per month (principal and interest), which you pay on the 1st of the month. Now, let's say that, in addition to your $1000 regular payment, you have an extra $1000 that you would like to pay to reduce your debt. So after you...
Last Sunday evening at the Beverly Hilton Hotel in Beverly Hills, California were the 67th Annual Golden Globe Awards hosted by English comedian Ricky Gervais. The choice of Gervais as host was a move to increase viewership which declined to 15 million viewers last year from 27 million in 2004. Canadian director James Cameron—who previously won in 1997 for his film Titanic—and his high-grossing 3D epic Avatar seemed to dominate...
Despite the prevalence of community detection algorithms, relatively less work has been done on understanding whether a network is indeed modular and how resilient the community structure is under perturbations. To address this issue, we propose a new vertex-based metric called "permanence", that can quantitatively give an estimate of the community-like structure of the network. The central idea of permanence is based on the observation that the strength of membership...
Questions people ask
Is the E5-small LoRA AI detector accurate?
No. It flagged 50.0% of human texts and caught 61.1% of AI texts in our test, a balanced accuracy of 55.6%.
What is the fastest AI detector you tested?
This one, at 0.03 seconds per 1,000 words on an Apple M4 Pro. Speed does not help when half of the human texts get flagged.