Ex-Berkeley PhD who leads SGLang at xAI explained how they serve Grok on 100K GPUs in 23 minutes - better than $2000 inference-at-scale courses.
split prefill and decode -> shard experts across GPUs -> route tokens per expert -> overlap comm and compute -> serve at DeepSeek-API-killing prices.
That loop is why xAI runs Grok on SGLang and third parties beat DeepSeek's own API by 5x on cost.
SGLang + prefill-decode disaggregation + expert parallelism + AMD MI300 - that's the stack.
Watch and save it, then read the article below.
The video of my Stanford CS25 guest lecture, From Language Models to Native Multimodal Intelligence, is now online.
I discussed how the core ideas behind LLMs has shaped multimodal AI, from architectures to training paradigms and scaling, and where the next challenges may lie. 🧠🌐
🎥: https://t.co/sUVxSEjrVf
Anthropic Engineer Andrej Karpathy:
"The biggest mistake in AI right now - people are forcing agents to work instead of mastering the model first
We made that mistake in 2016 at OpenAI - It cost us 5 years "
what Karpathy actually means:
step 1 → stop forcing your agent to do everything, understand the model underneath first
step 2 → demos are easy - products take a decade. self-driving proved it - if you skip the foundation, everything breaks
step 3 → the agent is not the product. the foundation is. build that - and agents emerge on their own
"you building agents right now - you're at the forefront. not OpenAI. not DeepMind. you "
watch - bookmark, then read article below ↓
New podcast with @garrytan, @farbood and Daniel Francis.
Live in the Future!
00:00 Guest Intros
02:35 Live in the Future
03:58 Will AI Outsmart us?
07:43 In the Anthropic Breadline
09:59 The Tech Genie Is Out
12:33 We Invested in COVID?!
14:25 Good Writing Is Novelty
18:50 Living Like It’s 2028
24:32 Truth dot ai
30:18 Does China have the Weights?
35:38 Everyone has AI Anxiety
39:32 Have Your Agent Talk to My Agent
42:01 What if Open Source takes the Lead?
44:03 The Sun is Setting on Google
48:00 Ride the AGI
50:46 Will There be Startups?
54:05 Defending Taiwan
1:00:05 The California Empire
1:01:26 If the U.S. Falls
1:03:11 Universal Basic Robot
1:06:01 Humans as AI Handlers
“We can take an open model and in the classified or non-classified context get it to the point of a frontier model. But you control the weights.”
各种长文总结,这句话才算要点
Etched has recruited legends from NVIDIA, Intel, Broadcom, Apple, and DeepMind:
- Brian Loiler: built NVIDIA's HGX and DGX systems from scratch over 22 years
- Wayne Cao: led 0-to-1 production for the original iPhone
- David Munday: built the TPU software team at Google
- Saptadeep Pal: NVIDIA's H100 architecture team
- Mark Ross: ex-CTO of Cypress (acquired for $9.4B)
They recruit through a process they call project-based recruiting:
"We map out all of the hardest technical problems across all industries that anyone has ever had to solve.
We look at who did the 0 to 1, who was actually in charge, who did the work.
We talked to as many people as possible, and then we just track it.
The amount of people who say yes after the first conversation is low, but the amount who say yes after the 20th is surprisingly high.
The number one guy in the world versus the number ten versus the number one hundred is a huge difference in whether it's actually possible to solve the problem."