Generative AI in Academia

Our group held a discussion on how Generative AI (GenAI) is shaping our work and research practices. Members shared their experiences using GenAI tools across different tasks and reflected on what works well, what doesn’t, and how we can use these tools more effectively. Most found it useful for improving productivity in coding, writing, learning, and administration, though views differed on how much to rely on it. Many also expressed feeling a need to adapt and make better use of GenAI, making this a timely discussion. Here are some of our key takeaways.

Coding

The overall sense was that GenAI is a powerful and useful tool for a range of coding tasks. Many agreed Copilot is useful for autocomplete, maintaining coding conventions, and speeding up function writing. ChatGPT also works well for plotting and visualisation, often more effectively than a standard web search. Opinions differed on how much to rely on these tools. Some felt GenAI should mainly be used to speed up tasks we already understand, because overuse could hinder learning. Others found it helpful for writing modular functions but stressed the importance of hand-writing tests to check they work as intended, especially for edge cases. Debugging with AI had mixed results - sometimes effective, sometimes misleading.

Scientific Writing

In contrast to coding, the group was more cautious in the use of GenAI for writing. Most agreed GenAI speeds up editing by fixing grammar and spelling. It can also be useful for finding synonyms and for rephrasing specific awkward sentences. However, group members were generally less comfortable with using it to turn notes into full sentences and to smooth entire paragraphs or documents: while time-saving, many felt it weakens the thinking process behind writing and can make outputs sound generic.

Learning

Again, the group expressed caution in the area of using GenAI for learning. Some found GenAI useful for summarising concepts and supporting quick literature reviews, though experiences varied in terms of accuracy. Some suggested that we test its ability to summarise our own papers before relying on it for new ones. Alternatively, we could use it to summarise a paper on which we hold critical opinions to see if it can identify similar problems with the study.

Admin and Organisation

Here the group felt positive about the benefits of GenAI. It can be helpful for planning, prioritising tasks, and drafting formal emails or event invitations. Some intend to use it more in this way, viewing it as a productivity boost with low risk.

Figure: Outcome of anonymous survey into how members of the Atmospheric Processes group use GenAI in their day to day work

Laura Mansfield
Hannah Christensen