A key early milestone in our journey to Superintelligence.

AI is expensive because the computers that train it are owned by a few big players. But thousands of GPUs around the world sit idle every day in gaming PCs, data centers, even universities. If we connect them into one decentralised network, anyone could rent that power for a fraction of today’s cost. That’s how we make AI affordable again.

I don't think that was ever in question but I doubt anyone expected it to become as commoditized though.

MyPOV - Vectors for Neural Networks. Not really a surprise. But also intrinsic design inefficiencies for speed and power.

In 2007, I offered to port the most CPU-intensive system to GPGPU at the quant fund I worked at. My boss told me that since I was the only security engineer, I should stay on that (kerberizing prod, SSO, automated patch mgmt, etc). Interesting fork in the road in hindsight...

Shit so good we created a bubble.

OG

Long-term alpha: AI academic papers

Visionary

I use CUDA Python to leverage NVIDIA GPUs for high-performance computing. This platform helps me accelerate computationally intensive tasks. It's great for large-scale projects. @NVIDIAAIDev #NVIDIA

Cathedrals everywhere for those with eyes to see!

GPU dominance wasn't just about floating-point ops; it was about parallelizing the massive attention mechanism. This architecture dictates the current scaling laws and model size limits for all large-scale, multi-step agentic systems.

!

Democratization of accessibility to powerful GPUs capable of running heavy AI workloads on the edge should be a game changer and something to watch out for.