#434 – Aravind Srinivas: Perplexity CEO on Future of AI, Search & the Internet
Lex Fridman Podcast
Decoupling Reasoning and Facts for Efficient AI Systems
The future of highly capable AI systems and exploring research directions rely on efficient inference compute, which requires a significant amount of GPUs. Winning in the AGI race is about having the biggest compute power, but breakthroughs can disrupt this by decoupling reasoning and facts. By representing knowledge in a more efficient and abstract manner, and making reasoning iterative and parameter decoupled, it is possible to create much smaller models that can reason effectively without the need for large clusters of GPUs.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
