A researcher argues that large language models exploit how humans naturally interpret meaning, highlighting the need for critical AI literacy grounded in understanding human interaction.

What happened: Mark Dingemanse, from Radboud University, has written a preprint chapter documenting how large language models—from historical precursors like divination and the ELIZA chatbot to today's systems—work by outsourcing judgment to computational tools that present fluid, confident-sounding output. The chapter is part of an edited volume on critical AI studies and is currently under review.

Why it matters: The ease and appeal of these systems risk masking the interactional and interpretive work happening behind the scenes. Dingemanse argues that their design—featuring fine-tuned overconfidence and interactive presentation—exploits the natural human processes we use to make sense of language, which means understanding human interaction and sense-making is foundational to developing genuine critical literacy around AI.

What to watch: The preprint is currently under review and likely to be revised, with the author explicitly welcoming comments. This work frames critical AI literacy not as a technical skill but as a social and linguistic one.