Are you annoyed with the anthropomorphizing language being used in the "AI" discourse, but not sure how to talk about this stuff without it?

[michelpatrice@jasette.facil.services]

@emilymbender

Peut-être davantage blessant que méprisant.

[ramin_hal9001@fe.disroot.org]

> artificial intelligence → probabilistic automation
hybrid intelligence → augmented human intelligence
image recognition → image labeling
speech recognition → automatic transcription
the model shows bias → the model reflects bias
model mistakes → model errors
chatbots are good at … → chatbots are good for …
hallucination → undesirable output

@emilymbender this is really great, thank you!

In conversations about AI, I rarely if ever hear people refer to a chatbot as anything other than "it," maybe I keep good company, but the output of these machines is still so unnatural to me I just can't see these things as conscious, or even just cognisant.

I do, however, frequently take shortcuts in talking about LLMs when saying things like, "trying to get it to understand," or "arguing with it," or "it thinks that..." I know I should say things like "trying to bias the output" instead of "trying to get it to understand," but the linguistic shortcut of using the word "understand" is just so much easier sometimes.

[majick@mefi.social]

@Imprinted @emilymbender It's okay to not be an expert, but... Prof. Bender *is*.

[dmakarios@theres.life]

@emilymbender I'm no expert, but how about "baseless output" for "hallucination"?
Even non-"hallucinated" output can be undesirable for various reasons, but the key point about "hallucinated" output is surely that it is groundless.

[emilymbender@dair-community.social]

@DMakarios the thing is that it is all groundless with LLMs.

[fluidlogic@oldbytes.space]

@JamesWidman @emilymbender him and his organs.

[andrew_shadura@mastodon.social]

@emilymbender
I guess this sentence is easy to misread as "avoid he/she, use they". It took me two or three attempts to read it the right way. Maybe if the last bit "as a start" was moved to the front of the sentence, it'd be easier to read?

@MichelPatrice

[lydiaconwell@todon.nl]

@emilymbender This is brilliant. Great for sharing to educate people.

I don't think 'probabilistic automation' however is a term that is likely to catch on and using the term is likely to make people think you're talking about something else.

I think 'intelligence simulator; might be a better term. It is probably not accurate way to describe AI and I would defer to your knowledge on this, but to me it could be a term people would understand as another way of saying AI, yet the 'simulator' part is slightly more of a put down than 'artificial.' Also the emphasis on simulator rather than intelligence.

[laslab@waag.social]

@emilymbender
In my (limited) understanding of English I associated "prompt" only with computers, so decided to check the Cambridge dictionary and found this to my surprise.
"... that has some of the qualities that a human brain has ..."
#LLM #language
https://dictionary.cambridge.org/dictionary/english/prompt

[labspokane@mas.to]

@JamesWidman @wordsmith @emilymbender Non-deterministic bilge engine