Dylan Patel

SemiAnalysis
Boutique AI Infrastructure Research and Consulting
DMs are open for consulting, quotes, or to talk shop,
Opinions my own

[Dylan Patel @dylan522p]

@creatine_cycle "Dinner"

[Dwarkesh Patel @dwarkesh_sp]

Cmon, this is unfair. Dylan's had a job before #sexworkiswork

[Dylan Patel @dylan522p]

Love that all FinTwit people making fun of Leopold still get mogged by his returns this year.
Leopold is a great guy and he's extremely in tune with where the world is headed.

[Dylan Patel @dylan522p]

@MollySOShea Jesus Christ lol

[Dylan Patel @dylan522p]

@RandianCapital That wasn’t my job bro lmfao
You can do things without it being ur job

[Dylan Patel @dylan522p]

@luke_drago_ Ty for posting this

[Dylan Patel @dylan522p]

@lillysharples @jaiyagill @Madisonkanna @SemiAnalysis_ It's Dwarkesh book, has quotes of lotta awesome people!

[jaiya @jaiyagill]

on a mission to find the coolest people, companies & ideas in tech, money & culture...

so naturally i had to see why @dylan522p, founder of @SemiAnalysis_, is so optimistic about AI

turns out the future has more jobs and less "permanent underclass" than doomers think

[Dylan Patel @dylan522p]

@MichaelSte5069 @SemiAnalysis_ You’re gonna love our next newsletter :)
Unfortunately it’s hard to justify selling it to individuals, B2B is the only way we can justify the immense effort we have in it

[Dylan Patel @dylan522p]

@BronsonSchoen I read the letter and personally know dozens who signed this, and discussed with them IRL.
The letter brings no solution on how to do this or even attempts to

[Dylan Patel @dylan522p]

Slowing down AI is a ultimately wishful thinking
The genie is out of the bottle
The most importan competition ever with potentially immeasurable benefits to the winner will not suddenly have people stop and sing kumbaya
While many lab employees have signed, many more havent.
Practically none at some labs less than 6 months behind have.

[Anthropic @AnthropicAI]

We support this petition, signed by our CEO, several co-founders, and senior staff.

Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field. https://t.co/DqwuQfa9xH

[Dylan Patel @dylan522p]

@AkashDwivedi5 They will. Nothing will slow down the progress meaningfully. Especially not petitions that don't care to understand China's role here

[Dylan Patel @dylan522p]

@Madisonkanna @aifjs @Reshusaur @creatine_cycle This is worse than corgi cafe

[Dylan Patel @dylan522p]

I would wait in line for 8 hours to watch Shrek 2 in 70mm IMAX

[Dylan Patel @dylan522p]

@growing_daniel @typedfemale whats the bit?

[Dylan Patel @dylan522p]

@ahmetbuilds 🫡

[Dylan Patel @dylan522p]

@typedfemale

[Dylan Patel @dylan522p]

@SQQQbear Shitposting is fun. Note this is the only tweet I've done that's pure AI output for a reason lol

[Dylan Patel @dylan522p]

Many don't know this but the collapse of the Bronze Age is what led humans to developing test-time compute, also known as an internal monologue.
When civilization collapsed, writing systems disappeared across much of the eastern Mediterranean. Trade networks broke. Specialized bureaucracies vanished. Knowledge could no longer be offloaded to institutions.
Humans increasingly had to simulate, plan, and reason locally rather than rely on civilization as external memory. We replaced external embeddings with internal compute.
The next transition may be the reverse: civilization increasingly becomes our external reasoning engine, while humans specialize in asking better questions.

[Dylan Patel @dylan522p]

Impressive that Huawei shipped this first after Nvidia / Groq and AMD / Cerebras announcements.

[vLLM @vllm_project]

🎉 Congrats to the teams behind vLLM AFD Plugin (Ascend & vLLM, @StepFun_ai, @AntGroup, FastAFD): a new experimental plugin under vllm-project that brings Attention-FFN Disaggregation to MoE serving.

Attention and the expert/FFN path are two very different workloads that normally share one topology. AFD runs them as separate services, so you can scale attention and experts independently. Same vLLM serving surface, no fork. @NVIDIA GPU and Ascend NPU.

[Dylan Patel @dylan522p]

@willdepue @LuisOala Limits have obv changed a lot around, but yes. Most people don't hit limits quick or have it flip over to API spend quickly tho