Ⲕⲁⳙ⳨⳨ⲩ: ᴀ Kᴀᴜꜰꜰʏ ɪᴍᴘʀɪɴᴛ
@kauffy.us
790 followers 780 following 5.7K posts
𝚝𝚑𝚛𝚎𝚎 𝚝𝚘𝚍𝚍𝚕𝚎𝚛𝚜 𝚜𝚝𝚊𝚌𝚔𝚎𝚍 𝚞𝚙 𝚒𝚗 𝚊 𝚝𝚛𝚎𝚗𝚌𝚑. dev. gamedev. ai. solo. neuroinsurgent.
Posts Media Videos Starter Packs
Pinned
kauffy.us
It's kind of sad to think that there's now a generation of young people for whom the word Kennedy refers to RFK Jr, rather than Jamie.
kauffy.us
I have a cassette somewhere that has this song on it four or five times-- each time, started late, or the DJ talked over part of it, or whatever.
kauffy.us
Philosophically, if you believe in a world that essentially exists in language, these models are the first "artificial minds"-- even if they only exist during the moments after you press Enter. To understand a bit better, read about the Turing Test and The Chinese Room (~thought experiments).
kauffy.us
.. training it in other ways to give it a pleasant personality. In the case of image models, they work the same way, except they are exposed to millions of pictures which _also_ have descriptions with them. After seeing enough pictures of a cat that include "cat", it now knows what a cat looks like.
kauffy.us
.. which is exposing them to enormous, enormous amounts of data. For a language model (an "LLM"-- like ChatGPT), they have it read billions of words. From this, it learns what conversations are, infers rules of grammar, etc. The creator then may "tune" it to be better suited for a purpose, like..
kauffy.us
.. that match what the prompt (etc.) was looking for.
It really is just "predicting the next thing" (in words' case)-- but it's predicting the next thing from the field of all possible things. The smaller the field is, the less varied the possibilities.
How models are created is by "training" them..
kauffy.us
What the model represents is a "possibility field"-- that is, it's possible to make any sentence or any picture from a given model.
How this works is using an algorithm that starts with your input ("prompt"), and uses this to traverse the field "collecting" (i.e., outputting) the bits of the model..
kauffy.us
I liked reading it, but laypeople are not going to understand that.

I've tried whittling it down to:
All the generative AI-- AI which outputs stuff from our input-- that exist now primarily rely on a model: a huge set of data that "surmises" all the elements of the domain of the model, words or pic
kauffy.us
My guess would be it's not a rule or law-- because, in theory, the Governor could go nuts and have "his" (or "her") National Guard troops doing something stupid, so the Federal Government, which is supposed to represent all the other States, would likely have power to prevent that.
kauffy.us
Fine. If they want to tear up the Constitution and make up their own insecure eurosterone nation, they better start saving now-- we bought all that shit.
kauffy.us
I think *not actually doing* a fascism by busying the resources that a fascist would use and removing that fascist's excuse to do it is fairly elegant. I mean, if it's necessary to "mobilize the national guard" (to do nothing), why not have the Governor do it properly?
kauffy.us
Nah. Newsom is prolly doing "a something" so Trump can't pretend to justify his wiener-compensating bullshit.
kauffy.us
I've done 3/4 of those things. The past few weeks have been the closest to doing it again. And settling for 3/4.
kauffy.us
Do you think pardoning Maxwell will be the way of drawing MAGA's fire so they continue to not notice the "gamines" thing?
kauffy.us
My first thought was "A Jewish girl named Katie?"
kauffy.us
That seems like decades ago. So quaint.

Slight correction: he stole funds to not build the wall.
kauffy.us
Power corrupts. Absolute power corrupts absolutely.
kauffy.us
I've argued this point with my Dad, who brags about the effectiveness of Israel's military. Yeah-- we pay for that with our healthcare.
kauffy.us
Sounds like Trump isn't getting very good information.