Latent Space: The AI Engineer Podcast

RWKV: Reinventing RNNs for the Transformer Era — with Eugene Cheah of UIlicious

15 snips
Aug 30, 2023
In this discussion, Eugene Cheah, CTO of UIlicious and a key contributor to the RWKV project, dives into the revolutionary RWKV model. He explains how it sidesteps traditional Transformers, achieving superior efficiency and context handling. The conversation highlights the significance of community-driven AI resources and how RWKV addresses memory limitations in processing large datasets. Cheah also explores the balance between open-source licensing and the use of coding models in enterprise settings, showcasing the global shift in AI technology.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Scaling AI on Large Websites

  • Eugene's company built an AI to analyze websites and suggest changes.
  • It failed on Amazon's website due to its massive HTML size, exceeding the model's token limit.
INSIGHT

Linear vs. Quadratic Scaling

  • Transformers recompute everything for each token, leading to quadratic scaling.
  • RWKV scales linearly because it only needs the current token and state.
ADVICE

Finding Hidden Gems in AI

  • To discover alternative models like RWKV before publicity, explore non-traditional sources.
  • Look beyond mainstream AI research and explore diverse communities and forums.
Get the Snipd Podcast app to discover more snips from this episode
Get the app