Max Leibman2d ago
Look, I know AI is controversial, but just for a moment, let's set aside our preconceived notions, our biases, the environmental impact, the massive cost to train and run models, the labor exploitation, the intellectual property theft, the inaccuracies, the mania it causes in users, the destruction of search, the deskilling of professionals, the devaluation of creative work, job losses, and lack of economic value from enterprise implementations.

Wait, what were we talking about?
Show threadtuban_muzuru

@maxleibman

Sigh. I sure wish the Chicken Littles had taken a little more linear algebra.

Just saying.

Jun 25 at 7:35pmWeb
Show threadMagnus Ahltorp2d ago
@tuban_muzuru @maxleibman How much linear algebra do you need to see the light, because I’ve taken quite a lot of it, and I ain’t seeing it.
Show threadtuban_muzuru2d ago

@ahltorp @maxleibman

It does furnish the working vocabulary and terms of art

Training AI models usually involves optimization, which means finding the best set of parameters (weights and biases) to minimize an error function. This process heavily relies on concepts like gradients (derivatives of multi-variable functions) and iterative updates, all of which are intrinsically linked to linear algebraic principles.

Show threadtuban_muzuru2d ago

@ahltorp @maxleibman

I'll just say it: linear algebra becomes indispensable for the computations that drive AI.

Matrix multiplication are central to neural networks, where input data is transformed through layers of weights to produce outputs.

Concepts like eigenvalues and eigenvectors are crucial in dimensionality reduction techniques (e.g., PCA) that simplify complex datasets, making them more manageable for algorithms.

Show threadMania Emma2d ago
@tuban_muzuru @ahltorp @maxleibman yes. This is theory. Only theory.
Show threadtuban_muzuru2d ago

@Emmaf_77 @ahltorp @maxleibman

Simply put, without the vocabulary of linear algebra, including all the calculus of loss and error functions, gradient descent, even backprop is the chain rule - if you can't do the math, do everyone a favor and don't embarrass yourself by saying dumb things.

Stats and Probability fill in for noisy data, inferential statistics, combinatorics, optimization theory.

But linear algebra is the vocabulary of these things.

Show threadkaneel1d ago
@tuban_muzuru @Emmaf_77 @ahltorp @maxleibman I'm trying to figure what is your point here… coz it sure sounds like you think all of the aforementioned points by OP are all not very important if we can do fucking cool nerd shits (that don't really work well anyway).
Show threadMagnus Ahltorp2d ago
@tuban_muzuru @maxleibman “derivatives of multi-variable functions” is calculus, not linear algebra, but that is only relevant when you know the function and can make a derivative, which means that you have to go to yet another mathematical field: numerical analysis. All of which I have studied (including machine learning), but it still doesn’t answer the question why fancy chatbots are considered a magical entity that is predestined to solve all problems.
Show threadtuban_muzuru1d ago

@ahltorp @maxleibman

… and who might think they can intelligently discuss AI _without_ linear algebra? Riddle me that, Magnus.

Show threadMagnus Ahltorp1d ago

@tuban_muzuru @maxleibman Then what is “AI” to you? Is it limited to machine learning that is linear algebra based? Neither “AI” nor “machine learning” is necessarily based in linear algebra, so you only seem to only mean “AI” in a post-deep-ANN sense.

And studying the effects of fancy chatbots can be done perfectly well without knowing one iota of linear algebra, which I know partly because I know linear algebra and partly because I know things besides linear algebra.

Show threadNot🐧A🐧Convicted🐧Felon5h ago
@ahltorp @tuban_muzuru@ohai.social @maxleibman If you can still stand up, you haven't taken enough.
Show threadMagnus Ahltorp4h ago
@sleepyfox @maxleibman I have since risen.