AI-Generated Text Detector

Overview

Code for fine-tuning a BERT model to detect AI-generated text. The resulting classifier labels a piece of text as either student-written or AI-generated.

What it does

• A Jupyter notebook (berttttt.ipynb) that trains the model end to end and runs predictions.
• A pre-trained model and tokenizer (bert_finetuned_model, bert_tokenizer) stored via Git LFS, so the classifier can be used without re-running training.
• A small prediction helper: tokenize the input (truncated/padded to a maximum of 512 tokens), run the model without gradients, and take the argmax over logits, mapping 0 → student and 1 → ai.

The README shows two example predictions: a sentence about blockchain "revolutionizing education" is labelled ai (1), and "Let us eat together any burger." is labelled student (0).

How it's built

• Model: BertForSequenceClassification from Hugging Face Transformers, fine-tuned for binary classification
• Tokenizer: BertTokenizer
• Framework: PyTorch
• Environment: Python 3.12, Jupyter Notebook; a CUDA-enabled GPU is optional but recommended for faster training
• Distribution: Git LFS for the fine-tuned weights and tokenizer

Results

The README does not report evaluation metrics; only the two qualitative examples above are given.