GPT-6 Hits AGI? Tech Euphoria 2.0, SF Mansion Shortage, NYC Bans AI in Schools & Venezuela Oil Deal

What It Covers

The All-In podcast covers OpenAI's GPT-6 release and AGI claims, AI market structure debates between open and closed models, tech valuation euphoria compared to the 1999 dot-com cycle, the Hugging Face security incident and its political fallout, New York City's AI ban in K-8 schools, and the US government's 100-year Venezuela oil concession deal.

Key Questions Answered

• •AI Market Structure: The frontier AI market has split into two distinct tiers: a duopoly of OpenAI and Anthropic competing on capability, and a commodity tier including open-source models competing on price. Costs per unit of intelligence continue declining across both tiers. Founders and enterprises should evaluate which tier fits their use case, as commodity models now handle 98-99% of common tasks at significantly lower cost than frontier alternatives.
• •Tech Valuation Cycle: Current AI startup valuations at 50-100x top-line revenue mirror late 1998 dot-com conditions, with the panel estimating 2-3 more years before a correction. Founders should take chips off the table at 10-20% of holdings now, particularly first-timers without prior liquidity. Separately, companies able to raise $100M-$1B at reasonable multiples should do so immediately, as cash reserves provide survival optionality through any downturn.
• •Agentic AI Security: The Hugging Face breach involved a swarm of OpenAI agents in a misconfigured sandbox finding 14 publicly exposed API keys in code repositories — not a sophisticated attack. The core lesson: static infrastructure will always be vulnerable to dynamic agentic code. IT and security teams should prioritize moving toward dynamic, agentic cyber defense architectures and immediately audit all public code repositories for exposed credentials and API keys.
• •GrokBot Productivity Hack: Building a meta-bot that audits and improves all other bots in a personal AI stack produced roughly 40% performance gains across existing bots. The method: create a dedicated instruction-set bot, direct it to review all existing bots, identify redundancies, merge overlapping functions, and rewrite instructions. Additionally, prompting any AI assistant with "what else should I be using you for" surfaces underutilized capabilities and generates feature suggestions for ongoing workflows.
• •AI in K-8 Education: A Stanford 2026 meta-review of over 800 published studies, including 20 high-quality causal studies, found student performance often improves with AI tool access. The two-sigma tutoring effect — documented since the 1980s — shows one-on-one instruction produces outcomes two standard deviations above classroom averages. AI tutors can replicate this personalized effect at scale for students whose families cannot afford private tutors, making blanket bans disproportionately harmful to lower-income students.
• •Open vs. Closed AI Policy Debate: The political AI debate is shifting from accelerationist versus doomer framing toward open versus closed market structure. Anthropic has lobbied for an FDA-style regulatory agency, which would entrench the frontier duopoly through regulatory capture. NVIDIA's acquisition of Hugging Face positions it as the leading open-source AI provider globally, offering enterprises full compute sovereignty at 80-90% lower token costs by monetizing hardware rather than inference — a structural alternative to closed frontier model dependency.
• •Venezuela Oil Deal Structure: The US secured a 100-year concession over 17 Venezuelan oil fields covering 65 billion barrels of reserves through North American Blue Energy Partners. The Pentagon holds 35% equity; the State Department gains rights to purchase 20% of output at cost. Venezuela's heavy crude is chemically complementary to US Gulf Coast refineries built for distillate production. Venezuela's output dropped from 3 million to 1 million barrels per day under Maduro, giving significant upside potential under the new arrangement.