This is a naive geometric approximation that doesn't take into account error correction. LLM's are often better critics than generators.

Is he just ignoring the test time compute papers and self correcting chains of thought?

Emad finishing his Batman arc.

What’s the best spot to cowork in SF right now? I want to grind Saturday

Deep Tech Week starts Monday in SF 65 events and thousands of people building technologies of abundance that elevate the human condition from suffering and fear, to abundance. Come for the talks. Stay for the party. See the Future. Deep Tech Week

Claude Code sub agents are wild. The swarms have arrived.

LLMs are crowded. GNNs for atoms? Just getting started ⚛️ UMA 1.1 is out — a universal MLIP for real-world chemistry, biology, and materials. High-accuracy. Open-source. Huge open field. It feels like AlexNet days. One GPU, one idea = real impact that can change the world!🌎

The Library of Almosts

Whispers of the Dream Tree In twilight's hush where fireflies roam, A tree of visions finds its home. Its branches rise through drifting mist, With mirrored leaves the stars have kissed. Each leaf reflects a tale unspun, A pirate ship, a rising sun— A crescent moon in oceans

I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in order to align with that, on a fresh Grok 4 chat with no custom instructions. grok.com/share/c2hhcmQt…

Incredible work!

I’m soo psyched to announce this one. We have raised again! Raised the bar for AI inference. By snapshotting GPU memory, we can cold start containers running vLLM or many other things extremely fast. This will be a game changer for many customers.

I've been running a GPU swarm with Agent mode. It's amazing.

Dawn of the Digital Age

GPT-5 is a sick model. It's hard to explain, but real world usage for code it really shines. It's much smarter than o3. I'm glad they didn't optimize just for benchmarks. It's fast, smart, and capable of much longer outputs than earlier reasoning models.