The most important point to remember is to avoid using any confidential information in unlicensed or non-secure versions of AI technology. This includes personally identifiable information about students, student works, or research data. Data anonymization may reduce risks, but it’s not a universal solution for avoiding the exposure of confidential data. Even with Microsoft Copilot, we're not completely sure of privacy-friendly treatment of our data during processing, as SURF actively advises against the use of Copilot. More information is available from the Dutch Data Protection Authority.

As with many online tools: if it’s free to use, you’re the product. For AI tools, this means all data is not only stored and sold, but also used to train future models. However, there are options to run LLMs and other AI models locally. While the capabilities may not be as advanced as those of commercial platforms, they still provide a similar experience in a privacy-friendly way.

In classroom settings, using LLMs with educational content is challenging because it may not be feasible to have all students install local LLMs. At present, the UT does not have a privacy-friendly contract with a cloud-based vendor. Therefore, these tools can be used by students in an inspirational way, but students cannot be required to log in and use their data.

Did you know? You can also run a Large Language Model (LLM) on your own machine using tools like Ollama and LM Studio, and models with a lower number of parameters can run on regular UT devices.