I’ve been writing linear algebra code for the WSE-3 since November and agree that Cerebras doesn’t need CUDA.
Most CUDA functions don’t even make sense in the WSE’s 2D programming model.
AI as we know it might not exist if @nvidia had not gone all-in more than a decade ago.
CUDA was foundational to the development of AI and remains important for training.
Inference, however, is different.
CUDA has no role in inference.
AI is now useful.
For coding. For doing your taxes.
For finance, legal and HR departments.
People are coding their own apps, eliminating SaaS applications.
And none of them know anything about CUDA.
And of course it’s this way.
It always is.
When a new technology emerges the practitioners use low level software.
It looks like math.
Its written by the select few.
Graduate students.
Supercompute engineers.
But to get popular, the technology needs to leave this environment.
There simply aren’t enough technologists to make a market.
There are no low level programming languages that are popular.
They are just too hard.
In order to get popular, computer languages move from low level to high level.
Compare how many people write Assembly versus Python.
It’s: 1 vs. 10,000.
When a technology is used by those who know nothing about its creation, and instead care only about its usefulness, then it has broken through.
It takes 12 keystrokes to link a world class AI model to your application, and start getting blazing fast tokens.
You don’t need to know anything about AI.
There is no CUDA.
Just point to the OpenAI API. Or the Cerebras API. Or to Fireworks. Or to DataBricks.
Here is what you type to move from Nvidia GPUs to Cerebras.
API key= os.environ.get(“CEREBRAS_API_KEY”)
That’s it.
@skeptrune@0xajka@theo@michael_chomsky theo’s right, you should do all the editing yourself. you’ll have an easier time recording because you’ll know what you can and can’t clean up while editing.
it’s also completely fake.
“Curious about its accuracy, I decided to stake out my local Domino’s. The plan is as follows: I’m going to place an order for delivery while hanging out in the restaurant, then compare each progress mark on the tracker with what’s actually happening with my pizza. Then I’m going to follow the delivery guy in my car until he arrives at my house, where my wife will be ready to receive him. For the record, I’m going into this hoping that the pizza tracker is not in fact a lie”
39 startups are building alongside Vercel for six weeks.
Backed by $8M in credits from AWS, Anthropic, OpenAI, ElevenLabs, and more, they have the resources to scale their infrastructure.
Meet the 2026 Vercel AI Accelerator cohort.
https://t.co/3H8v7xtdP5
if you’re not getting stuck anymore you’re not trying hard enough.
pick something 100x more ambitious than anything you’ve ever worked on and just start. you will get stuck. you might get un-stuck and build something great.
I realized something else AI has changed about coding: you don't get stuck anymore.
Programming used to be punctuated by episodes of extreme frustration, when a tricky bug ground things to a halt. That doesn't happen anymore.
@caballerobrah@SMT_Solvers I've never used SPARK/Ada but I built a SQL parser and query optimizer in Agda and ended up writing a lot of proofs along the way.
frama-c looks cool. have you seen clang's experimental bounds safety flag? https://t.co/Lo6SOJEmCA