Inference, Diffusion, World Models, and More | YC Paper Club
Categories: VC, Startup, Design
Summary
YC's new Paper Club brings together elite AI researchers and founders to bridge the gap between cutting-edge ML research and startup building. A Stanford researcher reveals why inference speed—not just training—will determine AI capability in the next 1-3 years, fundamentally shifting how founders should think about model deployment.
Key Takeaways
- Inference costs now dominate training costs at scale (billions of users = trillions of tokens), and RL-based training is actually just inference wrapped in a loop—making inference optimization critical infrastructure, not a secondary concern.
- Inference speed will become a capability lever, not just a cost lever: algorithms where performance scales with thinking time mean that tokens-per-second directly determines peak intelligence deliverable to users within latency constraints.
- Geographic arbitrage opportunity: ~50% of Bay Area AI talent is in San Francisco proper (Anthropic, OpenAI, Cursor), while significant talent cluster exists in Palo Alto (Google DeepMind, Tesla, XAI, startups) without deep YC integration—creating pool of untapped founder-researcher connections.
- Winter 2016 YC batch (140 companies) produced 10-15 unicorns while OpenAI co-founders were still involved in mentorship conversations—demonstrating value of embedding research leaders in startup selection and early guidance.
- Speculative decoding (and inference optimization broadly) requires systems-level thinking beyond model training—token multiplication matrices hide significant algorithmic and systems complexity that founders building AI products need specialized expertise to optimize.
Related topics
Transcript Excerpt
All right. Hello everyone. How you guys doing? Welcome to the first ever YC paper club. This is like a very exciting thing. Absolutely thrilled with the response. We had over a thousand folks that applied to come in. It was a very hard selection. If you guys have friends that didn't make the cut, I'm very sorry. We're we kind of we need to keep it to about a hundred. Um and so we selected a very very cool group. Um the mission is to create this kind of community of great founders and great researchers and try to pull them together. I guess just for you guys to get a sense for how cool the people in this room are. Um, raise your hand if you have at least five citations, 10 citations, a 100 citations, a thousand citations. Wow, this is insane. Okay, 10,000 citations. Oh my god. Okay. All rig…