Peter Devlin

@mund0sinfin Me reading about heterogenous compute approaches to inference.

Am I the only one that gets confused that KV Cache is stored on HBM? Or maybe it's my view that cache implies SRAM that is flawed...

@evanbuhler I keep getting wires crossed when I see like "KV" and "L1" in the same publications and pitches

@AlectryonTech Damn, might as well be written in Klingon for me…

@neilzegh Actual conversation between my highschool CS teacher and I when my solution to a maze routing problem was bogosort...

Me defending my O(n^3) solution to the coding interviewer.

@andre_slav03 This is awesome!

Wrote a fused mHC layer in pytorch using the kernels as a backend, and achieved some interesting speedups! Tested on H100 SXM5

@viemccoy Congratulations!!!!

THE HEIR TO THE THRONE IS BEING SUMMONED INTO EXISTENCE THROUGH THE MOST DIVINE PORTAL IMAGINABLE: MY WIFE! im so happy. im having a baby with the love of my life. this is the most important thing ive ever done.💠

@viemccoy youtu.be/iOk6HB609po?si…

Just letting you guys know I completed the system of German idealism but Im keeping it a secret so you have to work harder. Its kind of not fair for me to just release it

@viemccoy I wonder if the greater divergence from human thought is the fundamental latent space (type of mind, portals), or if it's more the process (gradient descent go brrr), or both.

Reasoning models aren't actually thinking - they're doing something closer to quick and efficient navigation of latent space towards a basin that is close to the type-of-mind that can answer your question

@lukeshiels Can’t wait

we're cooking something incredible

@lukeshiels @interfere_ Los Alamos energy

@interfere_ team hq (my living room)

@jia_seed Check out the SPH

any builders in zurich?

@sudoredj 🔥

geopolitically relevant chip conference uses stock powerpoint flame animation lol

@hthieblot Filter out 90% of the advice that you are given. Nobody cares as much as you, nobody knows as much as you

You can give a first-time founder just one piece of advice. what is it?

@katedeyneka @lilianweng Same! Highly recommend.

2. Lilian Weng @lilianweng - my journey into diffusion models started with Lil’Log. lilianweng.github.io

The fastest way to get smarter is to read smarter people. I went through @tbpn’s Metis List and pulled together blogs from some of the most influential people in AI. I’ve also added a few gems outside the list 🧵 I find the most value in blogs where people share applied insights and digest SOTA research. Share your favorite AI blogs in the comments!

@hanna_haya You gotta love the second longest fjord in Norway.

@viemccoy How does it compare to Kimi in terms of capabilities?

I like gpt-5! it's a good model, with a reasonably interesting personality, and good capabilities. It was not the massive discrete jump everyone was hoping for, but I think its a fitting model for the name (all things considered).

@monty10x @josefchen 🫡

so hyped to be speaking alongside @josefchen & @AlectryonTech this week. come along lu.ma/7glbnbbc

🛠️ 🇬🇧🧪

@gracegerwe @hanna_haya no contest

Who are the best female deep tech VCs

@arlanr Have you tried Kimi?

It's been only three days 💀 thanks opus <3