Anthony on Nostr: I'd offer the following observations: The fact that people motivated by vast amounts ...

I'd offer the following observations:

The fact that people motivated by vast amounts of money and power to criticize the term "stochastic parrots" choose to criticize the term "stochastic parrots" is not evidence that there is a problem with the term or the critique it implies. If someone has a critique, they should put it forward and defend it. Otherwise, pointing to "lots of people critique this" sounds at best like bad faith insinuation. Bender and collaborators have also put forward [the octopus test](https://aclanthology.org/2020.acl-main.463.pdf ), if one is feeling chauvinistic about the relative intelligence of birds.

The substance of Searl'e Chinese Room objection has never really been met by proponents of the belief that sequence-to-sequence maps somehow encase "intelligence" or "agency" or other cognitive markers humans have. Ken Aizawa and collaborators have written several nice articles on "the mark of the cognitive", trying to get at what we mean when we say humans are cognitive, and an immediate consequence of their work is that LLMs are not. By my read they don't go so far as to claim things like LLMs *cannot* be cognitive, but personally I think their work strongly implies that conclusion (they don't want to pick that particular fight is how I read them). A lot of these "debates" turn on the fact that to claim LLMs do something "cognitive", you must define precisely what you meany by that word, yet proponents of this view rarely if ever do.

In other words, and to speak of credibility, the burden of proof lies with the proponents of LLM "capabilities". Extraordinary claims demand extraordinary evidence, etc. Worse still, we've known about the ELIZA effect for 50-ish years by now, and since it is a powerful confounder it needs to be ruled out at a bare minimum before any further conclusions can reasonably be drawn.