Jason Wei delivers a talk on understanding AI in 2025, focusing on three key ideas: intelligence becoming a commodity, where the cost of knowledge and reasoning approaches zero due to adaptive compute; Verifier's Law, stating AI's ability to master tasks is proportional to their verifiability; and the jagged edge of intelligence, highlighting the varied capability and improvement rates of AI across different tasks based on factors like whether they are digital, easy for humans, and data-abundant. He argues against a fast AI takeoff, suggesting instead a gradual, task-specific improvement, and emphasizes that fields like software development will be heavily accelerated by AI, while others, like hairdressing, may remain largely untouched.
Sign in to continue reading, translating and more.
Continue
