have been following along with the textbook + doing the labs for cmu's 15-213 (intro to computer systems) cmu's reputation is not unearned holy shit why is the first lab kicking my ass right now

post training team really focused on getting qwen2.5-7b-it to understand when it has absolutely no context and ask for more, haven't looked but wouldn't be surprised to find that they make fantastic agents for this very reason

ah neurips rebuttal (i haven't been outside in 24 hrs)

this is getting an insane amount of hate! holy fuck we are way too comfortable generalizing from limited information! you don’t know these people! they aren’t the evil asian girl you met at a party! or a cheating douche from high school! holy fuck

i feel very bad that i haven't achieved anything great™ by now, but i think greatness is mostly determined by a) uncontrollable factors b) the small actions you take consistently for a long period of time, so i guess all i can do is march on, and check back in at 26

new post. there's a lot in it. i suggest you check it out

aarch-64, aarch-64 why have you forsaken me? bro can someone at pytorch PLEASE update the wheels for pytorch 2.8 and it's deps to include aarch-64 PLEASE you don't understand what you've done

spot instances were not the move

> be me > interested in low-level optim of linalg kernels > decide to do weekend project optimizing some on school cs box > finish writing testing code + shit baseline for first kernel > compile w/ O3 to make things a challenge > compiler instantly achieves throughput bound

wow it seems like the public is just not understanding that MSL is a separate entity from the rest of their AI org.

🔍 Ever notice how attention layers only tweak the residual stream in a low-dimensional way? That low-rank writing is exactly why so many SAE features stay dead—until Active Subspace Init rescues them. 👇 #AI #ML #MechInterp

wow kimi is so much better at intelligent push back. kimi-reasoner will be insane

i've had to learn the hard way that there are two ways to go into a semi-elaborate program: 1. wing it and feel it out 2. think a lot b4 you code the rub is that 1 almost never works for advanced programs, and the process of rewriting over and over is just a noisy sim of 2