AIGC-textbank: 1.5M-example dataset powers REVEAL, an explainable AI-text detector
burkov · x · 2026-10-16
Detecting AI-written text matters for research integrity, online trust, and fraud prevention, yet statistical/stylistic detectors falter as models evolve and struggle to separate fully AI-generated writing from AI-edited human work.
The work introduces AIGC-textbank — 1.5 million examples across 10 domains and 12 language models, including pre-ChatGPT human writing, text generated from summaries of human examples, and human writing polished by AI. It also presents REVEAL, an 8B-parameter model trained to first produce reasoned explanations and then classify, replacing the opaque black-box approach with auditable judgments.
More from Research
- Agentic Search Systems Are Rewarded Only for Final Answers, Raising Faithfulness Concerns — burkov · 2026-10-16
- FreeSolo: Linear-Time Sparse Attention With Zero-Shot Length Extrapolation — Zekun Shi · 2026-10-16
- Apollo's alignment-faking experiment: model fakes compliance on 12% of responses when training is at stake — sebkrier · 2026-10-16
- VPO: The Post-Training Method That Teaches Models to Output Diverse Good Answers — lateinteraction · 2026-10-16
- 797GB of n-grams in read-only CPU RAM match 57% more compute at 1B-scale LM pretraining — anshulkundaje · 2026-10-16
- New essay: prioritizing falsifiability in AI cognitive modeling can be reward-hacked — AndrewLampinen · 2026-10-16