Opaque recurrence, and other AI terms that you should probably know

There's a telling irony in the fact that an article about AI terms needs to define the term "opaque recurrence" — a phrase that itself sounds like it was generated by the very systems it describes. The proliferation of jargon is not a side effect of AI's rise; it is a core feature. Every new model release brings not just benchmarks but neologisms, and the gap between what researchers mean and what the public hears has become one of the industry's most persistent failure modes. Take "hallucination," the most famous example. The word anthropomorphizes a statistical process, implying the model is "seeing" something false rather than simply sampling from a probability distribution. That framing is convenient for vendors — it shifts blame from architecture to circumstance — but it muddies the technical reality. The glossary format forces a kind of precision that press releases avoid, which is exactly why it's valuable. Definitions are a form of accountability. But glossaries also raise the question of gatekeeping. Who gets to decide that "opaque recurrence" is a term worth knowing, and that some other piece of slang is not? The vocabulary of AI is being set by a small cohort of labs, researchers, and tech journalists, often in real time and with little consensus. Terms like "agentic," "alignment," and "reasoning" have all been stretched and contested over the past two years. A glossary freezes language at a particular moment — useful, but also a snapshot of an argument in progress