I wish the lectures could just be as easy as all the other lectures from other modules aaaaaa
 Thalidomide Vintage Ad Shirt $22.14 |
 UFOs Are A Psyop Shirt $21.68 |
Thalidomide Vintage Ad Shirt $22.14 |
Why the FRICK is Machine Learning so hard
https://rentry.org/machine-learning-roadmap
you should skip:
- linear algebra (math shit no1 cares)
- calculoos (math shit no1 cares)
- probability (math shit no1 cares)
- python (github copilot does this)
and get right to Machine Learning Concepts
This looks useful, thanks!
I don't care about the programming, I just want to pass my exam.
Oh shit, so the course went over a lot more than your lectures did? Sounds great, might have to look into that :0
Ooooohhh! Its computers.
>you should skip linear algebra and probability
60 IQ gigamoron. Do us a service and have a nice day immediately
t. gatekeeper
t. skipping the foundations and fundamentals of how the systems works to become a glorified codemonkey who doesn't understand what they are working on and can only copy and paste existing solutions.
Nah, you're gatekeeping by posting shitty bait. You'll have to learn the math, no way around it
why would you skip the funny parts?
Bill Gates and Stallmann both aced Harvard's infamous Math55. A bad mathematician will never be a great programmer
bill gates is a cucked b***h boy and stallman is a literal who
>just skip math bro! You don’t need it for fricking machine learning
Lol lmao even 7/10 bait
How hard could it be? it's python. The equivalent of a game mod language. You channers are not as clever as you think you are.
>it's python
Spotted the midwit.
Yeah, sure, go ahead and then look like a moron when reading something like "There are different schemes / strategies in place to uniquely pick from the probabilities that the model created. For example, you can truncate and remove tokens based off their distance from the top token via Min P sampling.
Min P is something I proposed a while back that has seen adoption in different backends for hosting LLMs (llama.cpp, exllamav2, text-generation-webui's loaders, etc) which sets a minimum threshold for a token to "pass" for consideration relative to the top token probability."
That's just the glossary. I know I do look like a moron reading this because I lever learned statistics.
No I.python us a keyboard attached to the c or c++ actual coding for baby tier interface so to speak.
It is the interface to reuse existing libraries, indeed. I don't believe learning them covers ML, it's just using the black box abstraction.
Pytorch has rules that you fine tune like "weights" and "training" I might look into it more later but right now I have enough on my plate.
"for quality control, you can remove bad tokens with a technique i invented, minplatypus sampling
it's something i proposed a while back that has seen adoption in different backends for hosting language models, which sets a quality floor for tokens."
it's literally buzzwords
none of the things he said you should skip are even remotely relevant to your excerpt.
What you quoted is essentially a technique to distill prompts down into their essence, which is a necessary technique because context length is still an extremely important metric in available models.
Machine learning is not complicated, but it is deep. Math people are bad at communicating.
I trudged through a course on udemy, dude sounded like a fricking moron, took a practice test and only 30% of the content was in the lecture, i went back and looked. 4.5/5 star course.
Imagine how many rubber bands you fit on her.
What
UUOOOHHHH!!! IT'S COMPUTERS!!! TOT
I just didnt get what it was till now. I get now though. Computers. Got it.
Does anyone want to teach me some basics enough to pass an exam in a few days? I'll pay a hundred bucks if I pass btw
I'm going to sleep but in case anyone wants to take this job offer, pls drop ur Discord and I'll message ya once I wake up
Channers will claim they know c but not python pft
What's the exam format? Do you have any past papers? Not sure if I'll be of much help but can share some of my ML/DL notes.
ID: 106772167198248960
>a few days
Is this a midterm, a final, or just a random exam?
Regardless you need to talk to your Professor and your Advisor and possibly your Dean.
Ask yourself if you're too moronic at this subject to make it what you spend your life on.
It could be that the professor sucks; if everyone else is doing badly, then it's probably the professor.
If there's one student doing well and everyone else is shit, try to talk to that person in addition to all the other people I've mentioned. But don't be a moronic useless sperg (you obviously are, you fricking filth).
If you're dead set on doing this because it's your passion (obviously it isn't, or else you'd already have done the extra work; you'd BE that guy that gets a 100 on the test while everyone else has to get grade curve'd up) then you need to do a lot of extra work.
You need to basically put yourself through an entire second class during the hours you aren't at class, because either your professor sucks or you suck, but either way you need to do vastly more work.
If you just want to pass and move on with your life, hire a professional test cheat to take the test for you (if this isn't just an exam but actually a midterm / final).
>why are you telling me to talk to my professor and advisor and dean?
Because these people want to keep you on the hook, they want to keep you in school, they want to keep you paying for classes.
They want you to graduate with a degree, from their school, so that someone will see their school's name on your resume when they throw your resume in the garbage.
By the way, the only skill that matters in this life is being able to make friends with talented people.
Seriously and unironically, without reservation, it's all that matters. That's what 'networking' is, it's making friends with talented people. To do that obviously requires being nice and a bunch of other skills you don't have because you're a moron, but now you know.
>talk to your deen
>they want you to yadda yadda
>spacing
what a reddit tier response
I like Kilian Weinberger's videos.
Lmao I spent 4 days solving a math proof homework problem for the intro to neural networks course I took in college and got 0 points.
Yeah you most likely wrote a novel and presented no real mathematical arguments. You just have to get used to predicate logic
Nah it wasn't about predicate logic. It was the first time I ever saw partial matrix and vector derivatives and in the it didn't matter because the problem statement was over 5 pages and the prof was randomly changing index variable names on multi-line equations.
>oh that was an error you were supposed to ignore that
ecksdee
it's actually so ez just blindly apply shit and it magically works
it's ironic because anyone dicking around with hyperparameters always gets supremely dunked on by people making massive architectural changes
cargo culting is real, but let's not pretend that there are more than maybe 30 people in the entire world who really understand what's going on and how to do the next thing.
It's kind of like that in any field though no? Real innovation is in the hands of a very select(or sometimes just very lucky) few while the rest of us just work to the current standards. It's only because there's recently been so much new attention and research being conducted that everyone's constantly scrambling to keep up.
Take anything I say with a grain of salt though because I'm not even a software engineer, just some control engineer student who stumbled onto a MSc course on intelligent systems.
At least with software engineering I can safely disagree.
The reason real innovation does not happen in software engineering is simply because of the amount that you have to accomplish to be innovative.
The ladder gets pulled from over you, so to speak.
For example, Intel Optane could have been a revolutionary technology that would bring back OSes that just don't reboot. It never happened because you would have to write an entire OS and port all of the software everyone wants to use on it.
Everyone uses web shit to make applications these days because GUIs in C++ or Rust are a complete nightmare. No one wants to make a GUI framework (except Microsoft) because you have to support 3 operating systems that intentionally do weird shit out of pure spite
It just keeps going.
Has there been anything of value since transformers? That's where I fell off the train and (from afar) it seems everything has been a refinement of the process and on scaling up.
All I know about transformers is google made the paper then they made bard then they made gemini. So it doesn’t sound like transformers is relevant
>tfw MSc in CS and got promoted to data scientist at my company but now I have to actually learn ML which I only encountered tangentially in support of computer vision stuff I did in university
Guess it's time to put my math hat back on and see how hard learning past the age of 30 is gonna be.
For the math:
https://people.math.harvard.edu/~shlomo/
Sternberg's Advanced Calculus is freely available online, for those who've wanted it but are unaware
>Why the FRICK is Machine Learning so hard
It's not really hard but it was worked out by mathemathicians and mathemathicians are smug entitled narcissistic homosexuals who like to explain everything in the least understandable way. They always use the least used words and greek letters to confuse you.
If you don't understand anything just search it on youtube and add the word "visualization" to your search. There are a shitton of really good channels with good animations what can explain NN-s to you.
I honestly expected some fricked up NN video where she pulls out and chews other's tongue.
> I wish they told me what I already know, so that I don't have to think about it
What kinda machine are ya learnin, Westinghouse?