my opinions on machine learning and ai stuff

june 13 2024

so, i got into studying data science at Berkeley in 2018. here's why i think ai and ml have had their perception stunted.

when i hear ai or ml from the news, content creators, and even job descriptions, theres a feeling of magic! which yeah! there's plenty of Actual Magic. like, these millions of tokens and features getting turned into vectors with so many dimensions and those get transformed with parameters and hyperparameters only meticulously computed by machine! BERT and GPT in their pretraining form use unsupervised learning, which means theyre fed tons of unlabeled data to which they come up with their own features for determining their predictions. it's just incredible how well they can work despite us not being able to understand those features that make them so effective, and that goes for the ocean of unsupervised models.

even still, everything depends on the *existing* data you give it.

it's not the kind of magic that can think and create on its own. it's not the kind of magic that swipes away the millenia of human storytelling, philosophical thought, real world inventions and implied society.

please, world, i want you to learn how models work. please don't be afraid, i want to show you that you can understand a pipeline and even make your own with a little help. data extraction, data labeling, cleaning, transformations; i think everyone can grasp that, but can i show people the limits of classical computing, invisible and ineffable features that embed bias, and the amazing power of your creative mind?

companionship- when a language model can "remember" you and give you words of comfort or advice for coping, i think that's wonderful. but the ability to sense what is not effable between people is like

红豆生南国，
春来发几枝，
愿君多采撷，
此物最相思。Wang Wei

see with your eyes, say it with the air only you exhale, hear it with your ears, and picture what is only your conceptualization.

feel it in your heart? i mean, it'll be a different sense for everyone, but you made that, and no one else can replicate that. language models... we want them to replicate!

so it's getting late and i think i'll stop here for now. this kinda rant has not been like, the why ai and ml have come to be like this, but why i think people need to realize the outputs we are making from it are not as world-changing as people can be, with how they think and all. more to this so i can review what i learned while also being interested in human cognition and specifically language acquisition.