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

[kevinashworth@mastodon.social]

@emilymbender
You had me at probabilistic, unverified software manipulator.

[toadjaune@hostux.social]

@emilymbender now I so want the same resource in French.

I've been struggling with exactly this

[emilymbender@dair-community.social]

@toadjaune Je crois que les mêmes stratégies doivent fonctionner assez bien.

[emilymbender@dair-community.social]

@kevinashworth PUSMic FTW!

[strange_new_words@tenforward.social]

@emilymbender I did an AI training yesterday and the trainer referred to his various Copilot agents as "he" and "she." Would that be considered at all anthropomorphizing language?

[emilymbender@dair-community.social]

@strange_new_words yes, as described in the linked post under names and pronouns.

[strange_new_words@tenforward.social]

@emilymbender Yes, I'm reading it now. Stupid me for assuming anything would be too flagrant to be included.

[kelson@notes.kvibber.com]

@iamnotU @emilymbender To be fair, I've been naming my computers for decades now

[iamnotu@mastodon.social]

@kelson @emilymbender I've been calling my computer names for decades now

[scriptfanix@maly.io]

@emilymbender "synthetic text extruding machine" is a mouthful, but is also very funny yet very fitting

[imprinted@mastodon.uno]

@emilymbender
Call it LLM, because "AI" is a marketing proposition.

[emilymbender@dair-community.social]

@Imprinted Wow -- did you reply without reading the newsletter in the post you're responding to or did you read it and decide that you needed to share this wisdom with me?

[imprinted@mastodon.uno]

@emilymbender
I read it and I agree with what you wrote. I'm saying that if we stick to the LLM definition, automatically there is no intelligence. But I'm no expert.

[martinvermeer@fediscience.org]

@laslab @emilymbender @Petesmom @Mimesatwork

> factuality violation has a connotation that all other output is factual right

But that is the case, even if only by accident!

No, my reservation is that it suggests a formal error condition, like zerodivide or floating-point overflow, when in reality it is just a variant of normal operation. You cannot even know it happened until you, as a human, introduce external information...

[stitam@fosstodon.org]

@emilymbender

Super interesting topic, thank you. For wider adoption I think it would be desirable to find alternatives that are also neutral in their sentiment. I spent some time thinking about "agentic AI" where the proposed alternative was "probabilistic, unverified software manipulator" which I think is very far from neutral. What do you think about "language-guided iterative solver"?

[mjj@mstdn.dk]

@emilymbender As we used to say, 'Knowledge wants to be free! But maybe it does not want to be anthropomorphized'

[michelpatrice@jasette.facil.services]

@emilymbender

« hallucination → undesirable output »

Why not a more straighforward "erroneous outpout"?

An undesirable result could a result that is accurate but that is not the one I am looking for. A made-up quote is more than undesirable, it is erroneous.

[michelpatrice@jasette.facil.services]

@emilymbender

« Pronouns are chosen each time, and avoiding pronouns usually reserved for people (and pets), e.g. he, she, and singular they is a good first step. »

Isn't singular they "it"?