Add HandPenned detector submission - #201
russellsailors-hub wants to merge 2 commits into
Conversation
Two DeBERTa-v3-large judges trained on 29k human / 29k machine paragraphs across 23 kinds of writing plus a slice of RAID train; score is the mean of the two. Predictions for all 672,000 test rows.
|
Eval run succeeded! Link to run: link Here are the results of the submission(s): HandPennedRelease date: 2026-09-05 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 84.18 and a TPR of 91.33% at FPR=5% and 73.49% 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! |
|
Closing this — HandPenned has been discontinued and the project is no longer maintained, so the submission's links are no longer live. Apologies for the noise, and thanks for the work on RAID. |
Adds
leaderboard/submissions/HandPenned/withpredictions.json(all 672,000 test ids, score in [0,1], higher = more likely machine) andmetadata.json.Detector: HandPenned (https://handpenned.com/), two fine-tuned DeBERTa-v3-large classifiers voting; trained on ~29k human and ~29k machine paragraphs across 23 genres, including a stratified slice of the RAID train split (disjoint from any evaluation set we use). Contact: support@handpenned.com.
🤖 Generated with Claude Code