Evaluation

Leaderboard

You can find the current leaderboard here.

This leaderboard was calculated for a random sample of 20% of the test data for each genre, perfectly balanced for human data (0) and LLM-generated data (1).

Submission guidelines

You can make a submission by sending the required model files to pieter.fivez@uantwerpen.be

Any additional questions about the guidelines can be sent to the same e-mail address.

We will generate the predictions using this Colab script, which contains the necessary information about the required model files.

Here we host a simple example of a valid submission (with a dummy sklearn model) which runs successfully.

Please run your submission successfully in the Colab script before submission. For this you can use the development data, which has the same format as the test data.

Reference baselines

We've added some baselines for reference during training and development. 

These baselines can be found here, along with some explanation.