Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Explore how artificial intelligence systems develop internal representations of the world and how these models differ from human understanding in this theoretical computer science lecture. Delve into the fascinating question of what happens inside generative AI and large language models as they construct deep internal representations for their own use, examining both their powerful observable behaviors and hidden expertise. Investigate the crucial distinctions between AI systems' models of the world and human models, particularly focusing on how mismatches can lead to situations where systems inadvertently "set us up to fail" through our interactions with them. Analyze concrete examples from game-playing scenarios, such as what occurs when a chess-winning model is paired with a weaker partner, and geographic navigation challenges, including how route-finding models handle unexpected detours. Discover theoretical results that complicate our understanding by showing that successful generation can be achieved even by agents provably incapable of identifying the model they're generating from. Gain insights into the broader societal implications of these AI-human interaction dynamics and learn about cutting-edge research at the intersection of algorithms, networks, and large-scale social and information systems.
