The model was trained on a dataset of over 100,000 sentences, each labeled as either AI-generated or human-written. This approach allows the model to predict the nature of each individual sentence, which is particularly useful for highlighting AI-written content within larger texts.

Evaluation Metrics

The model achieved an accuracy of 90% on the validation & test set.

Usage

python

import torch
from transformers import BertTokenizer, BertForSequenceClassification

tokenizer = BertTokenizer.from_pretrained("shahxeebhassan/bert_base_ai_content_detector")
model = BertForSequenceClassification.from_pretrained("shahxeebhassan/bert_base_ai_content_detector")

inputs = tokenizer("Distance learning will not benefit students because the students are not able to develop as good of a relationship with their teachers.", return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits

probabilities = torch.softmax(logits, dim=1).cpu().numpy()

predicted_label = probabilities.argmax(axis=1)

print(f"Predicted label for the input text: {predicted_label[0]}")