Karawynn Long on the likely (very bad) implications of investment vulture firm Kohlberg Kravis Roberts buying ebook platform provider Overdrive.
Continuing the ebooks theme, author Jane Friedman has written a pretty comprehensive post about her experience with Amazon's ebook store not only selling books that appear to be created with generative SALAMI with her name on the cover, but they have also allowed them to be linked to her GoodReads author profile, signaling that she is indeed the author. When she complained, Amazon insisted she provide proof she has a trademark on her name – an absurd request.
This preprint from Perdue researchers suggests ChatGPT – which many people believe is particularly reliable and useful for coding questions – provided incorrect answers more than half the time. The interesting thing here, however, is that human participants in this study preferred the incorrect answers over correct ones due to the style they were written in and their length, which seemed to indicate accuracy and helpfulness.
This is a particularly important study in the context of the Mozilla Developer Network (MDN) recent decision to add an “AI helper” to its developer documentation site. MDN has long been considered the gold standard for reliability and accuracy in web development documentation – providing generated answers that are more likely to be wrong than not, where the audience is inexperienced developers, seems like a very bad idea.
Libraries and Learning Links of the Week is published every Monday by Hugh Rundle.
Subscribe by following
@email@example.com on the fediverse or sign up for email below.