Bender, five years on: "stochastic parrots" was about LLMs, not all AI

"Just a description," and only ever about text

Bender's core correction is about register, not detail. She rejects the idea that the phrase was ever an insult, calling it "just a description of what these systems actually are": machines that predict probable word sequences with no grasp of meaning. The misconception she keeps running into is a category error, where people take a claim scoped to language models and hear a claim about "AI" in general.

Does an LLM understand "four"?

The thread became the exact fight the phrase provokes. One reader argued from an older technology:

Lisp expert systems transparently don't understand the meaning of any symbols they process, yet with enough developer elbow grease they can do all the same things an LLM can do, with much higher reliability… AFAICT Opus does not properly understand concepts like 'four.'

— a commenter comparing LLMs to Lisp expert systems

Another pointed at the standard party tricks that expose the gap:

The standard rs in strawberry, should I walk or drive to the car wash, etc all play on the fact that they don't have any real world model or thoughts against which they can judge their output…

— one commenter

But the phrase itself has hardened into a shibboleth, one reader complained:

Nearly all (99%+) people who use this phrase are anti-AI and just looking to show off how much they dislike AI… So it's a great phrase because in just about every case I can ignore what someone says afterwards.

— one reader

A quieter coda came from inside Google at the time of Gebru's exit:

I was at Google when she quit, working in ML, and hadn't heard of her until the story broke… no one I spoke with knew about her, so I assumed it was brigading.

— an ex-Googler who worked in ML then

Five years on, the term outran its authors: coined to describe a mechanism, it now mostly marks a tribe.