Brilliant framing around the diffusion challenge. The memory bandwith bottlenck is really the hidden constraint that nobody talks about when discussing LLM scalability. What caught me is how this mirrors the cloud adoption curve you mentioned,where enterprises couldnt scale until they figured out everything-as-code. Seems like knowledge graphs provide that same kind of structural foundation,kinda like turning implicit organizational knowledge into something queryable and reliable.
James -- Definitely! The future likely isn't one solution, but a hybrid. Imagine KG-boosted agentic memory guiding next-gen models; that's the path to greater predictability
Brilliant framing around the diffusion challenge. The memory bandwith bottlenck is really the hidden constraint that nobody talks about when discussing LLM scalability. What caught me is how this mirrors the cloud adoption curve you mentioned,where enterprises couldnt scale until they figured out everything-as-code. Seems like knowledge graphs provide that same kind of structural foundation,kinda like turning implicit organizational knowledge into something queryable and reliable.
It’s always memory or I/o
Brilliant; while diffusion's timing is critical for GenAI, ensuring equitable acces and quality education for these new skills is equally vital.
James -- Definitely! The future likely isn't one solution, but a hybrid. Imagine KG-boosted agentic memory guiding next-gen models; that's the path to greater predictability
https://www.youtube.com/watch?v=Po6aHQpArq8&list=RDPo6aHQpArq8&start_radio=1