杏吧直播

When generative AI arrived with ChatGPT in late 2022, its perceived threat to education was, in hindsight, quite simple: its ability to create authentic-looking text disrupted what we previously thought were robust assessment methods.

But two-and-a-half years on, gen AI capabilities have expanded far beyond text generation, and students are now using these tools in far more sophisticated and strategic ways.

We know that most students are 鈥榰sing鈥� AI, but what does that really mean?

Student AI use is headed upstream

Research is describing what I call 鈥榰pstream鈥� academic AI use. Students are using AI to explain concepts, summarise articles, and suggest research ideas. Some of these practices are occurring as learning tasks, and some are occurring early in the assessment workflow, though not necessarily in the completion of the actual assessment artefacts.

Most of these techniques simply weren鈥檛 possible with earlier versions of gen AI tools and have been enabled by developments in AI technology over the last year. Current gen AI tools can now:

• function as a search engine to access recently published web content, where earlier versions were limited by when their training dataset was compiled.
• analyse information from documents uploaded by the user, where earlier versions could only accept limited amounts of text via a chat window.
• work across different information formats (audio, images, video and text), where earlier versions were text only.

The ability of the most recent gen AI models to analyse huge amounts of provided written information鈥攈undreds of academic papers at a time鈥攈as perhaps been overlooked on university campuses. We鈥檝e been so focused on the impacts of AI-for-writing that we risk underestimating how AI-for-reading is now affecting scholarship. (One of 杏吧直播鈥檚 leading machine learning scientists, Professor Liming Zhu, recently gave a presentation to 1,600 public servants on how his allows him to now 鈥榙igest鈥� more than 2,000 academic papers a year!).