Even the most sophisticated Large Language Models (LLMs) share a troubling flaw: they confidently invent information. This isn’t a minor glitch; it’s a fundamental weakness where AI models, seemingly out of nowhere, fabricate details and present them as fact. A response can begin with impeccable accuracy, citing sources and demonstrating understanding, then suddenly veer into falsehood.
The consequences can be startling, even dangerous. A recent example involved an AI suggesting glue as a pizza topping – a bizarre outcome stemming from misinterpreting online content. While some LLMs may be less prone to these “hallucinations” than others, no system is entirely immune, prompting the ubiquitous disclaimers warning users about potential inaccuracies.
Apple’s new Intelligence platform is not exempt from this issue. An initial feature offering notification summaries was quickly adjusted after the AI began misrepresenting news alerts. In a particularly alarming instance, a headline about a shooting suspect was incorrectly summarized to state the individual had died by suicide in jail.
The problem isn’t limited to misreporting facts; Apple Intelligence appears to be inventing entirely new words. A post on a popular online forum highlighted bizarre summaries from a weather app, describing “imbixtent light rain.” The term, though sounding plausible, is completely fabricated.
The original poster wasn’t alone in noticing this strange phenomenon. Other users chimed in, reporting instances of “flemulating” and “tranqued” appearing in their summaries. One user even observed “stricively” instead of “strictly” on multiple occasions, suggesting a pattern of linguistic invention.
A compelling theory suggests the AI attempts to condense information and, when unable to do so naturally, creates “portmanteau” words – blending sounds and syllables to fit the summary length. This “vibes-word” creation, as one commenter described it, seems to occur most frequently with the weather app’s summaries.
Currently, the extent of this issue remains unclear. Limited reports suggest it may affect only a small number of users, potentially tied to specific iOS versions, devices, or applications. However, the possibility of an AI subtly altering language raises intriguing questions about the reliability of these summaries.
If you utilize Apple Intelligence’s notification summaries, a simple check could help determine the scope of this issue. Have you encountered any unfamiliar or seemingly invented words in your summaries? Sharing your experiences could shed light on this peculiar behavior and help understand the boundaries of AI’s creative – and potentially misleading – capabilities.
