Downloads · 30 days
13
1% of all-time downloads
nothingiisreal/open-gpt-3.5-detector
open-gpt-3.5-detector is a text classification model from nothingiisreal. Use it when you need a label for a piece of text. It is set up for transformers. The card lists the license as other.
We have trained distil bert on this dataset [https://huggingface.co/datasets/nothingiisreal/HumanStories]
Downloads · 30 days
13
1% of all-time downloads
All-time downloads
1.4K
Public
Parameters
67M
268 MB on disk
Likes
4
Public
Click a slice to open those files.
.safetensors268 MB · 100%
From the Hugging Face model README
We have trained distil bert on this dataset [https://huggingface.co/datasets/nothingiisreal/Human_Stories]
It's kinda okay for sampling, but needs improvements and exposure to more synthetic data and types of mistakes LLMs do.
Overall I'm extremely impressed with how well this 68 million parameter model works, and extremely disappointed with how every single AI is getting picked up after only training BERT on GPT3.5 rows of the data.
Class label 0 means human, 1 means AI.
We tested these models all of which worked:
GPT3.5, 4, 4o
Claude Sonnet, Opus
Wizard LM 2
Gemini 1.5 Pro
It's really blatant how every single AI company is using the same watermark whether knowingly or unknowingly (through LLM "incest")