Dylan Patel

My favorite thing about the new college grads that I've hired is that they don't ask you how to do stuff They just put it in ChatGPT and fucking try even if wrong So much better than new grads asking you how to do shit without trying Chat is breeding agency into kids

@ali1749794 @dylan It's a shit post

Yo I heard if u press Up, Up, Down, Down, Left, Right, Left, Right, B, A in Sam Fransisco there's an infinite money glitch https://t.co/k50oEbhrMs

SemiAnalysis is hiring multiple cracked Data Engineers to build robust, non-shit pipelines for datacenters, fabs, AI etc If you have opinions on data topology + the way things "should be done" + still vibe code apply below 2 roles https://t.co/uEh5f1U0cL https://t.co/BUjucQutqg

So the Collosus 2 posts did fire in the US but on XHS they do poorly Which surprised me because Colossus 2 is so fucking cool But then I realized it's not impressive to build power or other industrial facilities fast or to build it across state lines to skirt regulations in China That's boring and normal Everyone needs to adopt the Colossus 2 mentality if we're to reindustrialize.

My life consists of slack and slop

Labubu's are the new stable coin Its over for crypto bros

So excited for Enfabrica x Nvidia I've been an investor + advisor in Enfabrica for 2+ years Nvidia strategy on CX9 + NVSwitch, while ahead of industry, could be better with different approach for fabric boundaries + KV Management Jensen recognized this imo https://t.co/yTClJcq69l

@JopeRoot87994 Yes. Custom HBM doesn't solve production capacity challenges

Everytime you think Elon has lost his touch or is falling, he does some crazy shit and ur like... Damn he really got it How Elon is building the largest datacenter in the world faster than anyone else is amazing Also what they are doing on the RL side is also equally nuts

Made an account on XHS and they don't believe its us :( https://t.co/MeFgFZd7fr

I keep asking my AI friends and they say there's this pre training and post training But whose actually doing the training Some talk about mid training too But where's the good training

I can't troll @_albertgu anymore about his architectures never making it to frontier models. Alibaba Qwen3-Next is using a hybrid model. So here we are celebrating. At least H-Net is a failure cc @main_horse https://t.co/1QJGFkDeJS

@Sino_Market Bro share price vs market cap... It's sti way lower market cap, just share price higher

@AcceleratedCS Shorter

NVIDIA has widened the gap for inference rack scale architecture yet again! Prefill specialized inference chips massively lower TCO per million input tokens on long context transformers As usual, other AI chip upstarts will follow this with prefill specialized chips, but later

Had a blast doing a Super Micro factory tour with Charles Liang the founder and CEO of SuperMicro. Saw GB300, B300, MI355X, and much more! https://t.co/N9B9l9wZxK

Damn :( https://t.co/FRuOyG9OeE

SemiAnalysis twitter is hacked crypto is fake and ponzi scheme don't click that shit

This is the coolest thing anyone has ever made of me Holy fuck As cocky as I am, I got imposter syndrome from this I'm not this cool wtf https://t.co/8qaAkFQska

ze euv must have ze ai le chat integration on EUV tools wen? https://t.co/ZlmyWBzpTM

@UncleSam303 @VideoCardz @nvidia We've brought up rtx pro and have had volumes and such for a while here https://t.co/iNAMUIHSWD

To all the inference providers out there https://t.co/Zl2zpXmnLZ

@when_to_fold_em What's the explanation of this jump? They count differently?

Since AMD's Advancing AI event, there is an net change of over 160+ newly disabled PyTorch unit tests on exclusively ROCm. There is a direct correlation between the the number of disabled/skipped test and the end user experience. We are glad that @AnushElangovan and his team is finally starting to prioritize this work for ROCm PyTorch.

@charles_irl @modal Amazing

@JordanSVIC @SemiAnalysis_ Ya in SF part of next week btw or in PA time to time too. Jon and I have a pod btw...

@robi_rahman @thinknonlinear1 Literally everyone who works on them...

@robi_rahman @thinknonlinear1 They aren't picked randomly wut. That's their internal names

AWS cloud revenue projections can actually be mapped if you follow they datacenter builds, accelerator purchases, and flow through to rental income of specific clients. Yippit can never do this.