[andrewnakas]

Adds a leaderboard submission for GPTTrace (Ettin-1B): leaderboard/submissions/gpttrace-ettin1b/ (predictions.json with all 672,000 test ids, plus metadata.json).

Model: jhu-clsp/ettin-encoder-1b (MIT), fine-tuned for binary human/machine classification for one epoch (bf16).

Training data:

• RAID train only (paraphrase-attacked rows oversampled ×2); no RAID test data was used;
• 200k rows of MAGE train;
• ~100k human texts from Wikipedia, PMC open access, arXiv abstracts and Project Gutenberg;
• ~16k topic-matched human/AI pairs with paraphrase/humanizer rewrites from open-weight models.

Inputs go through the same light normalisation as our earlier entry (gpttrace-ettin150): zero-width characters removed and Latin homoglyphs mapped back.

Scores: score is P(machine-generated) from the bf16 model.

Held-out RAID train titles (our own split): TPR@5%FPR 99.91% with attacks, AUROC 0.9995. This is the same research line as the 150M entry, a larger encoder plus more data.

[github-actions]

Eval run succeeded! Link to run: link

Here are the results of the submission(s):

GPTTrace (Ettin-1B)

Release date: 2026-10-15

I've committed detailed results of this detector's performance on the test set to this PR.

On the RAID dataset as a whole (aggregated across all generation models, domains, decoding strategies, repetition penalties, and adversarial attacks), it achieved an AUROC of 99.94 and a TPR of 99.90% at FPR=5% and 99.42% at FPR=1%.
Without adversarial attacks, it achieved AUROC of 99.97 and a TPR of 99.95% at FPR=5% and 99.77% at FPR=1%.

If all looks well, a maintainer will come by soon to merge this PR and your entry/entries will appear on the leaderboard. If you need to make any changes, feel free to push new commits to this PR. Thanks for submitting to RAID!

[liamdugan]

@andrewnakas let me know if you'd like me to merge this into the leaderboard

[andrewnakas]

Hell yes please merge it! Thanks Liam. I'm also committed to open sourcing all my ai detection work and going to continue to try to publish to absolute best ai detection technology possible. Here is a link to the model and weights on hugging face, which I just published and I think we updated the pr meta data with this info but yeah stoked to push this stuff forward! https://huggingface.co/nakas/gpttrace-v5-1b
My GitHub is updated with training, evaluation code and reproduction steps.
https://github.com/andrewnakas/gpttrace
gpttrace.com

Gotta get the number 1 spot one of these days. Also I'm distilling this model into the 150m model so hopefully that one will get a great performance boost and be able to run on webllm and consumer hardware well! Oh yeah!

[andrewnakas]

I think it triggered the second automatic test when I updated the meta data and such and I stopped it because it would just produce the same result, sorry about that one. Thanks for the help.

[github-actions]

Eval run succeeded! Link to run: link

Here are the results of the submission(s):

GPTTrace (Ettin-1B)

Release date: 2026-10-15

I've committed detailed results of this detector's performance on the test set to this PR.

On the RAID dataset as a whole (aggregated across all generation models, domains, decoding strategies, repetition penalties, and adversarial attacks), it achieved an AUROC of 99.94 and a TPR of 99.90% at FPR=5% and 99.42% at FPR=1%.
Without adversarial attacks, it achieved AUROC of 99.97 and a TPR of 99.95% at FPR=5% and 99.77% at FPR=1%.

If all looks well, a maintainer will come by soon to merge this PR and your entry/entries will appear on the leaderboard. If you need to make any changes, feel free to push new commits to this PR. Thanks for submitting to RAID!