A joint initiative of ETH Zurich and ETH Lausanne
As an instructor at ETH (ETHZ or EPFL), you can help build the next tools for supporting teaching and learning through AI.
We currently rely on commercial Large Language Models for tools like custom chatbots and grading assistance. These models have been designed to deal with a plethora of topics, but not always at the level at which we teach.
You can assist in building our own tuned Large Language Model, based on ETH Zurich and ETH Lausanne teaching materials. We welcome:
These materials would be used for tuning an open, pretrained model such as Llama2, Gemma, or Mistral, which we would then run locally on campus. The resulting model would hopefully speak our language of STEM education and be more closely aligned with our curricula, particularly if used in conjunction with other techniques like retrieval augmentation.
The model would only be available to ETH Zurich and ETH Lausanne.
By uploading materials, you confirm:
Unfortunately, the fair-use exception for teaching purposes does not apply to tuning a Large Language Model.
It is unlikely that the model would reproduce your materials verbatim, but they would irrevocably become part of the tuned weights of the model.
Thank you for your consideration!