Generative AI was not as good as humans at reading reams of documents on corporate regulation, turning in bland summaries that failed to pick up on nuances noticed by people, an experiment by the regulator found.
The Australian Securities and Investments Commission, in a pilot program as part of its work to prepare for the age of artificial intelligence, deployed a large language model to “read” and summarise public submissions to a parliamentary inquiry that referenced the regulator. It compared the results to humans who did the same task.
Loading...
Tess Bennett is a technology reporter with The Australian Financial Review, based in the Brisbane newsroom. She was previously the work & careers reporter. Connect with Tess on Twitter. Email Tess at tess.bennett@afr.com
