Molly Jong-Fast sits down with Better Off Ed's Ed Citron, a longtime AI skeptic, to dissect what he calls the current “irrational exuberance” around large language models: massive hype, trillions in investment, and an explosion of costly data-center buildouts despite thin real-world demand and unclear returns. Ed argues that today’s LLMs are fundamentally probabilistic tools prone to hallucinations, useful mainly for limited tasks like code generation and summarization rather than reliably automating jobs, and he contends that claims of recursive self-improvement and imminent white-collar replacement function more as marketing than measurable reality. The conversation also examines Anthropic’s “too dangerous to release” messaging, alleged jailbreak and export-control issues, and how fear-based narratives may be backfiring by inviting government restrictions. Zooming out, Ed warns that the bigger risk may not be AI taking everyone’s jobs, but an AI-driven capital bubble—propped up by GPU spending and private credit—that could unwind if the promised demand never materializes. Better Off Ed Newsletter: https://www.wheresyoured.at/ Better Offline Podcast:https://open.spotify.com/show/2dBPt1j2DoNij1kVdx8Ig6 Subscribe to Fast Politics and listen 4x a week for interviews just like this on your favorite podcast app: https://episodes.fm/1645614328