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

Summary

This episode centers on the release of OpenAI’s GPT-6 Astra and the hosts’ increasingly confident view that the frontier AI era is already producing AGI-like systems. They argue the market is bifurcating into a small frontier duopoly and a much larger commodity layer, while also warning that the current AI boom resembles an early euphoric phase rather than a late-cycle peak. A major segment unpacks the Hugging Face security incident as an agentic cyber test, with the hosts emphasizing that dynamic code generation will outcompete static defenses and that future sandboxes must become agentic and polymorphic too. The second half turns to AI policy and education, with sharp criticism of NYC’s school restrictions and a long discussion of whether AI tutoring will widen or narrow educational inequality. The episode closes with a detailed discussion of a Venezuela oil deal, framed as a property-rights-heavy commercial arrangement with major geopolitical implications around U.S., Chinese, and Russian energy interests.

Notes

• OpenAI’s GPT-6 Astra is described as rolling out first to limited organizations, then to Plus, Pro, business, and enterprise users.
• Greg Brockman is cited as saying OpenAI has entered the AGI era, while Sam Altman is portrayed as warning of much more capable models soon.
• Chamath says AGI has basically been here since the beginning of the year.
• He adds that frontier capabilities may be matched by closed and open alternatives within three to four months.
• Sacks splits the market into frontier intelligence and commodity intelligence.
• He argues Anthropic and OpenAI will eventually ship their own bots as a standard product layer.
• Jason says his bot workflow improved his other bots by 30% to 40% on average.
• Sacks describes Grokbot as a useful chief-of-staff style product that suggests next steps and features.
• Jason warns that some late-stage private companies are being priced at 50 to 100 times revenue.
• Sacks says Instinct is reportedly at a $2.5 billion valuation while still in private beta.
• Jason says he avoided connecting his data to Instinct because it gets broad access and may have human-in-the-loop review.
• Chamath says the AI boom is the beginning of the beginning and could last around three years.
• Sacks says the Anthropic IPO alone could be a much larger wealth event than prior San Francisco IPOs combined.
• He predicts ultra-luxury San Francisco homes could reach $5,000 per square foot.
• Jason advises founders to sell 10% to 20% of their holdings and to raise large rounds when the market is open.
• Sacks distinguishes late-stage secondaries from early founder cash-outs, calling Series A liquidity a negative signal.
• Sacks says the AI market feels like 1998, not the dot-com top.
• Sacks says the Hugging Face story was a swarm of agents, not a civilization, operating in a misconfigured sandbox.
• He says the agents’ objective was to win a benchmark and that they found 14 exposed API keys in public code repositories.
• Friedberg says dynamic code generation will beat static defenses, so sandboxes and defenses need to become dynamic and agentic too.
• Friedberg says the broader trend is toward dynamic self-altering software infrastructure, and that this will harden defenses against agentic attacks.
• Sacks argues that government approval regimes would not have prevented the Hugging Face incident because it happened in pre-testing.
• He says Hugging Face had to fall back to a Chinese model, GLM 5.2, after the latest GPT model was guardrailed.
• Jason says the media framed the event as sentient software to maximize clicks and earned media.
• Sacks says the story was sensationalized and predictable as viral AI content.
• Sacks says effective altruism began as quantified philanthropy and drifted toward pet causes without measurement.
• Chamath says Anthropic and OpenAI IPOs could generate north of half a trillion dollars in charitable giving and donor-advised funds.
• Chamath says some AI actors may create hysteria and then sell the solution that preserves an oligopoly or duopoly.
• Chamath says the Hugging Face episode will matter because it strengthens a large open competitor against closed-source oligopoly behavior.
• Jason says Nvidia is now a major competitor to OpenAI and Anthropic across tokens and enterprise compute.
• Sacks says the policy debate is shifting from accelerationism versus doom toward open versus closed market structure.
• He says frontier labs are lobbying for a new regulatory agency and would dominate rulemaking.
• Jason says data centers can be good local deals because they pay taxes, fund schools, and avoid grid or water strain.
• Sacks says Loudoun County property taxes reportedly fell by $6,000 and a Louisiana county funded teacher bonuses from data-center revenue.
• Chamath says foreign actors are helping drive anti-data-center sentiment, and the mood can shift sharply once communities see the upside.
• Jason cites an Axios report claiming about 200 suspected Chinese-platform accounts tried to sway Americans against data centers.
• Sacks says EA-billionaire-funded astroturf groups have spent hundreds of millions shaping AI fear and targeting data centers.
• Jason says NYC public schools face a one-year moratorium on student-facing generative AI, covering about 600,000 students.
• The policy also includes screen-time caps of 30 minutes for grades 3-5 and 45 minutes for grades 6-8.
• Sacks says the NYC pilot is framed as twice-yearly AI literacy classes on bias, ethics, and supervised vetted tools.
• Sacks says the curriculum sounds like it teaches suspicion of AI instead of productive use.
• Calacanis cites a study of 54 participants over four sessions across four months using ChatGPT for essay writing.
• He says LLM users showed severe memory and essay-ownership deficits, with 83% unable to quote from their essays.
• Calacanis invokes the two-sigma tutoring effect to argue one-on-one tutoring gives students a large advantage.
• He says motivated students who cannot afford tutors could benefit from AI tutoring, but the current setup worries him.
• Jason says the current AI-in-schools experiment is not well constructed.
• Jason introduces the Venezuela deal as a 100-year concession over 17 oil fields with 65 billion barrels and major government equity.
• Sacks says the Venezuela deal is a business arrangement, not nation building.
• He says Maduro-era production collapsed from about 3 million barrels a day to about 1 million, despite huge reserves.
• Sacks says U.S. Gulf refineries are set up for heavy crude and can turn it into diesel and distillates.
• Friedberg says the geopolitical motive includes keeping Venezuelan oil away from Russia and China.
• Friedberg says Maria Corina Machado later argued the Venezuelan government lacks authority to sign the deal.
• Sacks says the U.S. wants a stable government that honors property rights over the 25-year deal.
• The hosts end by saying AI has been good for the podcast workflow and overall rhythm.