with sp00ky season upon us, throwing it back to being in hell for this production of No Exit by ethan cavanac 😈
(don't @ me for using a stool; the show was 2.5 hrs with no intermission, and if i moved too much, i'd fall off the stage)
instagram.com/reel/DfIIPrbgB…
kimi K3.1 is just a post-train of K3? damn, was kinda hoping they'd make kernel engineers' lives more interesting
since it seems like many people share my indolence, who wants to have an "i was too lazy to go to san jose for pytorch con" meetup in SF wednesday night?
linear attention variants, though challenging, are really pretty and rewarding to work on kernel-wise. please keep them coming! (delta rule variants in particular)
the fact that you can't use both 2-cta and 1-cta tcgen05 mma instructions in the same kernel is truly one of the most baffling and frustrating limitations of blackwell
Reuben Stern retweeted
Tinkerers have been busy scaling up long-context RL! We’ve made significant improvements to Tinker’s efficiency to support those, and are passing these on with price cuts up to 70%.
GLM-5.3-Flash and DeepSeek-v4.1-Flash are also live for cost-efficient long-context work.
living the bicoastal life is ping-pong scheduling on a big scale
anyway, flying back to SF today -- who wants to make some music (SF orchestra book club!?) or talk kernels?
for reference, orchestra book club is my nonprofit orchestra that does sightreading parties
orchestrabookclub.org/
big fan of driss-bait
drisspg.github.io/nuggets/KD…
wrote up a new one; lil bit product pitch, lil bit click bait










