We're singularly focused on delivering 10 to 100s of GWs of compute faster than anyone else, rethinking every layer of the stack. Speed and scale are our key differentiators. Join @fluidstack to build civilization-scale infrastructure for AI. Software Engineer: fluidstack.io/jobs/software-… Product Engineer: fluidstack.io/jobs/product-e… Apply now or DM me.

SemiAnalysis x Fluidstack is kicking off GTC with a Power to Prefill, Dirt to Decode, Transformers to Transformers: A Full-Stack AI Infrastructure Hackathon. Build with the best. Bring your project, win prizes, and listen to insights from top industry leaders like @marksaroufim GPU MODE, @cHHillee Thinking Machines, Thomas Raoux OpenAI, @dylan522p, and more. Limited spots, apply below. You won't want to miss it. luma.com/SAxFSHack

SemiAnalysis x Fluidstack are kicking off GTC with A Full-Stack AI Infra GPU Hackathon Power to Prefill, Dirt to Decode Build with the best, win prizes, and hear @marksaroufim GPU MODE, @cHHillee Thinking Machines, Thomas Raoux OpenAI, @garywu Apply below luma.com/SAxFSHack

Incredibly excited for @fluidstack to partner with @AnthropicAI and help build a world of abundant intelligence. We are hiring - reach out if you want to work at the frontier!

Top tier swag game from the @fluidstack team🧢 look at the GB200 set!

jensen liked our gb200 legos

lego lithography means lego gpus as well

I wrote something about working at TSMC. Linked below.

Nicer video of Alfie cleaning an office.. next up, we’re working with Buck Lewis ( issuu.com/risd/docs/risdxyz_… ) to redesign Alfie to be more ✨🫧⚡️

An early timelapse video of Alfie, the general purpose teleop robot we've been building at @prosper Here it cleans a kitchen table, wipes the surface, rinses dishes in the sink before placing them in the dishwasher, throws trash, and puts other items away.

spent so many hours trying to get around context window limitations and now we have 32k. magical.

The context switching penalty when trying to debug an LLM chain 20 calls long is the worst I've ever experienced.

I actually hate prompt engineering its so unreliable and sensitive. "Ignore irrelevant information". "Think step by step". "Therefore the hypothesis is:" Why can't you just "Do what I want"

Technical moat for 95% of generative AI 'products' right now is zero. Most of these products are features that can be hacked together in a week. Market structure for these products will approach perfect competition.

As the field of machine learning advances, the alignment problem is becoming increasingly concrete. In a new report, I present the key high-level arguments for expecting AGI to be misaligned, grounded in technical details of deep learning. drive.google.com/file/d/1TsB… Thread:

"Language Models Can Teach Themselves to Program Better" This paper changed my thinking about what future langauge models will be good at, mostly in a really concerning way. Let's start with some context: [1/11]