Okay I had more thoughts and got more excited than I realised (because computer science is cool) so here goes
#it honestly depends on how pedantic you want to make your definitions#the definition changes HUGELY with context#for example#if you define 'AI' to be any non sentient thing that acts 'intelligently'#well then your washing machine is intelligent!#it can decide how much water is needed based on the amount of clothes#it can do you laundry differently based on the type of clothes#assuming you feed it the type of cycle as an input#sure but none of us like that definition. Even though very technically speaking you'd define automation as a subset of AI#we don't think of it like that#okay then cool. Now then there's the learning type. you'd say that's AI yes?#what about your keyboard? the one on your phone. or the one word apps used to have#they detect grammatical errors which#frankly#I myself may not catch#they can (with a good chance of being right) catch mispellings and give the right word#and they do learn#if you say hey this mispelling is intentional. boom. add word to dictionary. learnt#and the behavior changes accordingly#but oh nonono that doesn't count! you say#that's just statistics#well guess what LLMs are buddy#they're the best statistical models we can currently create of the human language#AND by modelling the human language they are able to 'reason'#an LLM has enough background data to say hey#a lion is to a lioness what a king is to a queen#that's automating the part where we write down and hardcode all those vague connections and keep updating the connections and numbers#as an aside this is actually so fucking cool!! the way the better we get at modelling the human language the better we get at mimicking US#and convincingly! in a way that the words being spit out are cohesive and meaningful!#i mean what is it exactly that makes us so different? The better we get at making things 'intelligent' the more we narrow our definitions
What I wanted to add here was - would you call a Tic-Tac-Toe playing program 'intelligent'? The intro to AI courses definitely do! But I think we just take it for granted that computers can do these things. That we can convert abstract math into physical circuits and 'tell' those circuits to do vastly various things. Back to the point - tic tac toe programs go through all possible plays, evaluate the opponent's possible plays and so on and so forth, evaluating each outcome's favourability in a very mathy way. We understand it and it makes sense to us - it's a process we explicitly define and we don't think of it as AI.
Okay diverting back to why I started typing so much in the first place - I see people say things like 'AI is evil' or talk about LLMs as if they are word generators that are incapable of anything. There are a couple of things here
LLMs are good at what they do and are frankly exceedingly fascinating
We WANT Computer scientists and engineers to do computer sciencing and engineering! Again, this is a very very interesting field
Yes, LLMs are not the same as automations and scripts we are used to
Yes, LLMs can be useful! Let's say you want to build a tool that alerts people if the web page they're on seems to contain a trigger they want to avoid. If I were building a solution, I'd probably use a language model. And probably a Large language model! Because Large Language Models are good at nebulous, language related tasks that would be a pain to try and even define clearly
But love the tool hate the wielder! Right now the industry is foaming at the mouth with a 'shiny new hammer can be used to fix everything ever. Smash 'em till they seem better'. And right now, the shiny new tech that corporations are dropping employees for functions to expose very clearly that
They've never cared much about workers over profits (revolutionary new knowledge)
They will make any excuses they can to 'optimise' the visible results while turning a blind eye to long term issues, reliablity implications, ethical implications, employee health and on and on and on and on
And also keep in mind that they profit from hyping this up as much as possible. They are creating and filling up the AI bubble.
Anyways my point is that people are focusing on the wrong thing!!! We should be driving towards fixing all these pre-existing issues instead of burning the tool! Every symptom that we talk about is a symptom of a fucked up society!!
Last couple of asides - I'm pissed about the fact that fields like 'AI Explainability' or other types of AI or even completely non AI fields are being swallowed whole into the LLM, genAI hype. Papers and tools just keep getting shittier because everyone wants to hop onto the next big thing that will get them money. This happened before, when an APPLICATION of blockchain (bitcoin & cryptocurrency - which is so fucking cool in its own right) got blown out of proportion to monetise the hell out, stunting what it could be!
Abstraction is everything! LMs are structs we have set up using mathematical constructs, putting pieces together to try to model language processes better (and we have been working on these for decades, based on how you want to define them). 'Regular' code and scripts are something we write that allows us to model a variety of processes (including language processes) using logical structs like 'if this then that' (which eventually funnel down into math anyway). At my level of abstraction, these are the same thing! What level of abstraction are you looking at this from?
Again, yes they are shitty! Because tech corporations are bypassing ethics, fiinding loopholes, blowing up the hype and rushing to push a new tool/model into doing things it may not be well equipped for! For money!!!! But what is it that bothers you? The capitalism or the tool it is misusing? Check with yourself!
germansingergirl
vaporwave
stormborns
yousharknotpass
ellievsbear
