Some takeaways from LPs at RAISE: - With everyone widening their mandate into AI, every pitch sounds identical. True portfolio concentratio...

Instinct seems to have completely lost mindshare in my network. Current split: @Muse for personal, @bot for work. https://twitter.com/ai/sta...

Increasingly hearing from friends & family how much they are enjoying Grok in their Teslas. Who would’ve thought Tesla would be the Trojan h...

AI inference is shifting from single GPU clusters to Heterogeneous Disaggregated Inference. Prefill (compute-bound) and Decode (memory-bandw...

Building a fast chip is hard. Running different chips together without slowdowns is even harder. https://twitter.com/tsait26/status/21105268...

Credit where it's due. @somi_synthefy had accurately predicted that frontier labs will be building their own github ~6 months ago. https://t...

GPU companies going upstream to the model layer. Hyperscalers & CSPs going downstream building custom silicon. Full stack, vertical integrat...

Slept on this so I didn't reply purely on visceral instinct because framed a certain way, calling seed checks "option bets" feels like a dir...

AI & agents are turning hardware into a search problem. https://twitter.com/kevingubbi/status/2106262345257554327

Turns out raw GPU token speed isn't the real bottleneck for agents. It’s CPU tool execution, memory bandwidth, and sandbox overhead. Interes...

Fascinating. Current LLM design is like "hardware Stockholm syndrome". Models were designed to fit CUDA warps and HBM limits as a compromis...

Kinda surprised Google has completely dropped the ball on their assistant app. What happened to Spark? https://twitter.com/google/status/205...

Powerful reminder that the AI wave isn't GPU-only. Agents have completely changed the shape of the workload. https://twitter.com/jukan05/sta...

In VC, we talk a lot about rate of learning over credentials. @NojanSheybani & @kevingubbi are a rare combo of both. We need founders like t...

Heard this framing the other night when thinking about reasoning models (like o1) vs classification/decision models (like Jev): system 2 mo...

Jev for high-level robotics orchestration & state machines feels like a prime use case. Fast schema outputs > LLM lag. Who’s building this?

This is the way! Retail banking has monetized customer inertia for decades with artificial friction. Agents will take that friction to zero...

This article nails why AI transition in silicon will be iterative. Chip design is a multi-objective search loop, not a linear flow. Great re...

Legacy infrastructure is not ready for these real-time agent onslaughts. It’s going to feel like being perpetually DDoS. https://twitter.com...

I’ve become increasingly convinced: the best startup founders are cult leaders. A founder's job comes down to constantly convincing 3 dist...