I recently read a good article on the different kinds of truth a language model operates with by Luke Mann, Liam Magee and Vanicka Arora, Truth Machines: Synthesizing Veracity in AI Language Models, but despite its lovely typology of truths (consensus, correspondence, […]
I have this hunch that sparkles and glow and shimmer are somehow a point in the latent spaces of LLMs that have more connections than you would expect. Perhaps their connotation to magic and to the unknown matches some of the mystique […]
This LLM-generated paper demonstrates how LLMs fail to cite core scholarship on a topic but generate misaligned citations systematically.
A genre glitch is a characteristic of LLM-assisted writing where the text suddenly switches genre, typically inserting a short promotional phrase full of sensory details into an informational text. Genre glitches occur when a word in the generated text is heavily associated […]
Grammarly is a full on AI plagiarism machine now, generating text, citations (often irrelevant), “humanizing” the text to avoid AI checkers and so on. If you’re an author or scholar, they also have been impersonating and offering “feedback” in your name. Until yesterday, when they discontinued the Expert Review feature due to a class action lawsuit. Here are screenshots of how it worked.
Readers who don’t read Norwegian: sorry. This is in Norwegian because it is commenting on a current debate in Norwegian media. Asle Tojes debattinnlegg «får KI-alarmen til å gå,» skriver Petter Bae Brandtzæg, og jeg er enig. Toje sier riktignok nei, han har ikke brukt […]