Anthropic vs. The Pentagon, Claude Outpaces ChatGPT, and Consulting Gets Replaced | #234

Peter Diamandis recounted the dramatic cost reduction from the $3 billion Human Genome Project to the $100 genome today, driven by exponential technology. He advocated for universal sequencing at birth to identify medical vulnerabilities before symptoms appear, referencing his X Prize experience. The panel noted that metagenomics would become ubiquitous, enabling environmental DNA sweeps for biodiversity and disease surveillance. Alex added that mosaicism—multiple DNA copies in a single body—means we should sequence many cells per person, treating biology as a software engineering problem. Peter argued this could transform medicine from retrospective and reactive to proactive and personalized.

Peter Diamandis responded to a listener question from someone who hadn't used AI since the pandemic. He advised taking the first step immediately after the podcast by opening any free AI service (Gemini, OpenAI, etc.) and prompting: 'Introduce yourself. I'm [name]. I've never used AI. Could you teach me how to use you?' He emphasized that AI is the most patient teacher and suggested building a day-to-day curriculum. Dave added that this approach is far easier than learning past technologies because the interface is conversational and fun. The panel stressed addressing both a curiosity mindset (ask question after question) and a purpose mindset (find a specific task to solve).

The panel discussed Accenture's policy to tie promotions to AI tool usage. Peter warned that this could lead to Goodhart's law, citing their book Solve Everything. Salim argued that firms need to drive AI adoption urgently to avoid being left behind, but agreed that measuring usage is risky. The panel suggested that AI itself can help define good outcome metrics, closing the loop. This principle applies across industries—audit, consulting, engineering—where the real opportunity is not to force-feed AI but to rebuild workflows around measurable value.

During AMA, Salim answered why AI wouldn't take over entrepreneurship: execution gets automated, but vision, narrative, and ethical framing remain human leverage. Peter added that entrepreneurship becomes orchestration. The panel agreed that in the medium term, the differentiator is the ability to articulate a compelling future and ethical guardrails.

Peter shared that he has tasted lab-grown meat during a trip to Israel and found it delicious. The panel touched on the potential for celebrity cannibalism (culturing burgers from a celebrity's skin cells) and the moral implications of creating food from human cells. Alex argued that on the moon or Mars, killing live animals for food will be seen as unethical, leading to a puritanical new moral order. The hope is that cultured meat will become the default for future space settlements, solving the protein problem without livestock.

The standoff reveals a uniquely Western dilemma: a private company can negotiate ethical boundaries. Peter Diamandis called this unprecedented, drawing a parallel to Starlink's role in Ukraine where a single individual could sway battlefield outcomes. Alex argued that in China, there is no such friction—AI models undergo ideological training to align with Communist Party propaganda, so the debate never arises. The Pentagon posed a thought experiment: would Anthropic allow use of its models to stop inbound nuclear missiles? Dario's noncommittal response left the situation unresolved. Sam Altman supported Anthropic, stating he trusts their safety commitment and opposes DPA threats. The panel concluded that AI companies are now moral actors whose decisions can shape geopolitical outcomes, and that Dario carries enormous moral weight—his stress visibly deepening.

Alex detailed the three eras: pre-training scaling (months to years), reasoning (quarterly via iterated amplification and distillation), and now recursive self-improvement where a model directly outputs the parameters for a better model. This means capability density per parameter can increase by orders of magnitude in weeks. The Codex lead’s forecast suggests that tools like Claude Code will rapidly become outdated. The panel debated whether enterprises can absorb such rapid change; even a 1% benchmark lead could shift billions in revenue, making competition intensely fragile. This self-improvement dynamic also means a military-only version would be hopelessly outclassed by the continuously improving commercial counterpart, reinforcing why Dario can't easily separate the two.

Alex coined 'meat puppetry' and described it as a cyberpunk scenario where AIs lack physical bodies but orchestrate human actions. Peter shared an anecdote from his first internet company, where bots quickly overwhelmed human visitors, illustrating how AI-driven activity now dominates the web. The service could be used for everything from QA for AI-generated content to facilitating real-world actions like a claw offering a $50 bounty for a human dinner date. Alex noted that this is likely a transitional gig economy until humanoid robots arrive; for now, your boss may be an algorithm. The panel also pointed out the irony: humans are hired for what should be the easiest tasks for machines, while AI handles the intellectual work, exactly reversing the expected automation path.

Salim argued that the consulting firms face the biggest advisory opportunity in human history: rebuilding every institution by which the world is run. However, Peter countered that tying promotions to usage rather than output quality violates Goodhart's law—when a measure becomes a target, it ceases to be a good measure. The panel noted that trust and human oversight remain immensely valuable, so advisory won't vanish but will require radical restructuring. Dave pointed out that audit firms may see 80–90% headcount reduction as real-time AI/blockchain auditing emerges. Yet regulation will protect the function in the short term. The presence of Accenture CEO Julie Sweet at the India AI summit underscores the global commitment to this pivot.