Overview
Generative AI models are rapidly becoming commonplace for a wide range of tasks, including those relevant to academic research. As a group that develops and uses AI models, we are in a unique position as early adopters of many of these tools. However, it is critical that they be used responsible.
Non-Generative AI Models
You are encouraged to use non-generative AI tools that will make you more productive, such as AI-based literature search tools. This section is primarily focused on generative models . Programming
Large language models like ChatGPT can be very helpful for programming, especially when using popular Python libraries and in combination with plugins like GitHub co-pilot. Their use is permitted, but they need to be used with care. Keep in mind that these models are not experts. They simply stitch together the most likely sequence of text. This means that they oftentimes produce code that looks right and may even run, but oftentimes the resulting code has subtle yet fatal flaws. So, you must always think through every step independently. These tools should be used primarily as a way to guide you towards the right solution, but you still need to know what you're doing. Ultimately, the burden of trust is on you to ensure that you use these models appropriately in your programming efforts.
Writing
In short, you are never permitted to use a generative AI model to write text for you for anything related to our research (e.g. papers, abstracts, reports). Writing should be in your own voice, with your own thoughts and structure. If you are struggling to write something, it is my job to help you, and I am happy to do so.
Images
Generative AI models should never be used on any data-related figures. For instance, they should not be used to enhance, modify, or make plots, as they can easily hallucinate in subtle ways. They can, however, be used for generating art used in a public presentation, but their use must be clearly acknowledged.
Data Privacy
Research data should not be uploaded to the server used by an AI model hosted by a third party. All research data is, strictly speaking, property of the university and/or that of a collaborator. There is no guarantee that data uploaded would remain private, and this can pose many challenges, especially in partnership with collaborators in industry. Chat with Andrew if you'd like to request an exception to this rule.
Research Models
Naturally, generative AI models are also present in our research field, such as diffusion models for crystal structure generation. Of course, their usage is allowed, and you should cite their use just like any other computational method employed in your work.