The creator of GPT doesn’t have a PhD. The creator of PyTorch doesn’t have a PhD. The research lead at Cursor dropped out of NEU. You don’t need a PhD or a top school to become a great researcher or engineer. You can just do things!

HAPPENING NOW! Jeffrey Katzenberg and @tomascooking in the HOUSE! Live on @tbpn

Today we announced Alembic Technologies' series B financing to accelerate our work in Causal AI. The round is a hefty $145M which is a 15.7x jump in valuation compared to last year’s Series A with Justin Wexler at WndrCo . A huge thank you to our lead Matt Roberts at Prysm

Running production with CUDA_LAUNCH_BLOCKING=1

So BF16 breaks RL and we should use FP16 instead, except it's actually just a problem with A100's so you're fine on newer hardware, but it's actually due to some arcane flash attention setting so you just need to check that, and otherwise we're probably fine with BF16?

I see a lot of people confused by the DGX Spark. The machine, to my understanding, was for those of us that have large DGX systems that need research development boxes. I need to match my infrastructure and have my science team not blow the breakers at the office. I don’t want to

Tokyo Disney architecture > Orange County. Orange County logistics and distances > Tokyo Rides are =

What a gorgeous place to have breakfast. #JapanVacation

Quote of the day: I certainly don't agree that machines which can solve IMO problems will be useful for mathematicians doing research, in the same way that when I arrived in Cambridge UK as an undergraduate clutching my IMO gold medal I was in no position to help any of the

So true. Windsurf and others are really bad examples of founders leaving their teams behind and not even sharing the proceeds with their team. I definitely would not work with their founders next time.

Product market fit in enterprise can be measured in number of mandatory founder red eye flights. 😭😭😭

The job of a startup is to find undiscovered talent and distill it into a product.

One of the interesting things with LLM tools when working on a Python & C++ / CUDA stack where you write the actual C++ and kernels. It automatically assumes that you don’t have access to the C++ and are using packages. This is even when you have it in the same repo. Since I

@jordihays As I feared. What an absolute betrayal of Silicon Valley principles and the common pool. The image it brings forth is that of a villain in a film. One that separates off the rear of a locomotive train that is full of civilians. Just so he can speed up the engine full of money

As we scale the number of pre 8 AM internal meetings compounds as it’s the only time that’s left.

With all the Soham memes going around I want to use the algo to remind people that we’re hiring on-site in San Francisco across pretty much every engineering role.

Watching @AIatMeta adding researchers is entertaining. Nice part about working on spiking neural networks, spatial-temporal dynamics, AI memory, and causal is that we don’t have to play in the transformer race. Evolving algorithms are going to win in the end.

Why can I not code atomic methods in CUDA without thinking of Eminence in Shadows “I AM ATOMIC!”