OpenAI's GPT-5 Flop, AI's Unlimited Market, China's Big Advantage, Rise in Socialism, Housing Crisis

Gavin Baker notes that on cutting‑edge benchmarks like Humanities Last Exam and RKGI‑2, xAI’s Grok‑4 slightly outperforms GPT‑5, a first for OpenAI after years of leapfrogging rivals shortly after they release frontier models. While OpenAI still tops some leaderboard-style tests and retains a massive product/distribution edge, the underwhelming keynote (flat/wrong charts, typos) and lack of decisive performance gains raise questions about brain drain, internal turmoil, and the difficulty of making step‑function improvements in LLMs.

Friedberg highlights that GPT‑5’s key product advance is automatic model selection: the system internally chooses between fast/light vs deep/reasoning models based on the query, sparing users from picking cryptic model names. His experiments—simple retrieval vs complex research—showed the router quickly serving easy questions and entering ‘thinking’ mode for harder ones. Gavin adds the router was buggy in the first 12–16 hours, suggesting performance may improve, but both emphasize that OpenAI’s product polish and distribution could sustain its value even if pure model leadership slips.

Baker stresses that big tech’s returns on invested capital have risen since ramping AI capex, unlike the early internet ‘dark fiber’ era. Meta’s AI‑driven ad targeting and engagement are boosting revenue, and Microsoft is seeing strong Copilot uptake despite product weaknesses. On the flip side, Anthropic alone is asking for 50 gigawatts of US power over three years—roughly 5% of US generation and equivalent to a full year’s added capacity—turning AI into the dominant driver of new generation projects and arguably a bigger long‑run energy catalyst than the Inflation Reduction Act.

The panel notes China is adding ~1 terawatt (1,000 GW) of power every ~18 months, dominates new solar additions, and is deploying >150 nuclear reactors. That scale, plus manufacturing cost advantages, supports EVs (BYD), TikTok, and even low‑cost chains like Luckin Coffee as tools of geo‑economic influence. Friedberg and Baker counter that post‑2018 crackdowns, VC winter, and the collapse in new company formation (Jack Ma’s ‘disappearance’, tutoring nationalizations) show the CCP throttling entrepreneurship; Phil Deutch adds that unpredictable contracts and no true rule of law undermine long-term innovation, even if short‑run corporatism looks impressive.

Ben Shapiro argues the AI race with China has a genuine existential dimension—both militarily and informationally (TikTok as psyop)—tempting the US to adopt Manhattan Project–style centralization. He and Gavin describe the “imperial presidency” rising as Congress abdicates to regulators and voters demand more federal solutions. Friedberg warns socialism will likely ‘sweep’ the US over time as half the workforce is directly/indirectly government‑paid, costs of living (groceries, rent) soar, and dependency grows. Both he and Shapiro emphasize that failed socialist policies rarely shrink the state; they typically invite “more government” to solve the problems created by earlier interventions.