forecast

Fluid intelligence: iterating on fixed weights

Once you crack the 'nature' part (pre-training), AGI becomes less about pre/post-training and more about fluid intelligence: applying the same model weights iteratively for a long time to verify facts, ask the right questions in a chain, and reason — this iterative inference (not just one-shot answers) is what needs compute and could produce answers worth a trillion dollars, like an AI inventing the transformer.

forecastpending

Speaker

Aravind Srinivas

Evidence Quote

“It’s fluid intelligence, right? The facts, research papers, existing facts about the world, ability to take that, verify what is correct and right, ask the right questions and do it in a chain.”

Source

Aravind Srinivas— Lex Fridman Podcast
Created: 6/13/2026, 3:26:05 AM

Your notes

Loading notes...