You Wanna Talk About Settler Colonialism? Talk About This:

The problem here isn’t that large language models hallucinate, lie, or misrepresent the world in some way. It’s that they are not designed to represent the world at all; instead, they are designed to convey convincing lines of text. So when they are provided with a database of some sort, they use this, in one way or another, to make their responses more convincing. But they are not in any real way attempting to convey or transmit the information in the database. As Chirag Shah and Emily Bender put it: “Nothing in the design of language models (whose training task is to predict words given context) is actually designed to handle arithmetic, temporal reasoning, etc. To the extent that they sometimes get the right answer to such questions is only because they happened to synthesize relevant strings out of what was in their training data. No reasoning is involved […] Similarly, language models are prone to making stuff up […] because they are not designed to express some underlying set of information in natural language; they are only manipulating the form of language” (Shah & Bender, 2022). These models aren’t designed to transmit information, so we shouldn’t be too surprised when their assertions turn out to be false.

ChatGPT is bullshit

tone tags are such an interesting phenomenon linguistically. bc they're of course intended to add clarity to text communication, in a way that's more explicit than nonverbal signals, but they're also more removed from the message than verbal signals are -- arguably more removed than nonverbal signals are too, bc they're so constructed.

and that distance gets in the way of efficacy, from my perspective -- like there's a reason we spontaneously developed emoticons extremely early on in the internet days (and eventually latched onto emojis so hard), and ~*~wAyS tO eVOkE tOnE~*~, and the ones that caught on are the iconic [in the linguistic sense, ie resembling the concepts they evoke] ones bc they can be guessed.

and i think there's also a reason that when speaking out loud we don't just say what we want to say and then add "genuine" or "not mean" at the end as if speaking it makes it so -- we use full sentences like "I'm really not trying to be mean here." that's the part that really interests me, the abstraction of concepts that have never been just about tone, have always needed carefully chosen words to be expressed at least sometimes. that's where i think tone tags are distinctly less effective than just writing out the sentence for someone to read as part of the message and not as an abstracted layer on top of the message. (whereas with stuff like /lh for lighthearted i kinda get it -- i still find a well-chosen emoticon or emoji does a much better job at making me read a message in a lighthearted tone, but i get the desire to be explicit.)

ik this reads like a critique even though i started out saying it was just interesting, and idk it's both. i think the desire to implement a completely explicit tone layer on top of messages is fascinating, and i think its failure states are also fascinating, and im not about to grump at people for using an imperfect tool bc they're all imperfect.

Anyone want a gem tutorial? ^^

If you want to support me in making more of these: https://www.patreon.com/PixelArtJourney

this is problematic of me (joke) but i really enjoy the splashing of french into english speech or writing. just adds a pizzazz

“Absolute freedom mocks at justice. Absolute justice denies freedom. To be fruitful, the two ideas must find their limits in each other.”

— Albert Camus, The Rebel

†

TiL (click to go to the thread, which probably has more interesting tidbits I missed).

Bonus: