Sam Altman’s Attack, Amazon vs. Starlink, and What Opus 4.7 Actually Means | #248

Alex noted that despite virtual tools, displacement exists because people are unwilling to move, and mobility maximizes options. Dave pointed out that San Francisco's AI scene is concentrated in Mission Bay. Salem cited Steve Blank's research: most Silicon Valley founders came from elsewhere, because hometowns punish audacity. Peter shared Philip Rosedale's LinkedIn analysis finding the Bay Area has 10x the entrepreneur density of anywhere else. The group mentioned the VC who funded a founder through seven failures because they knew he wouldn't stop—a culture that cannot exist without density. They recommended explicitly choosing your environment to shape your outcomes.

Dave referenced Drew Houston's MIT commencement speech from 2017, which argued that scientific studies prove you become the average of the five people you spend the most time with. He said this is a positive for the podcast team but stressed that it should not be left to chance: 'Choose those five people very very carefully. Don't let it just default to random. Choose them explicitly.' This protocol ties to the earlier geographic mobility advice—moving to the right city is a way to choose a better default set of five.

Salem explained that the Luddites weren't against machines, they were against the owners who kept all the profit. He suggested that as AI collapses hierarchies, profit sharing becomes essential. Peter described Robert Goldberg's model: he goes into mid-market engineering, construction, and trucking firms, and the first thing they do is implement profit sharing, which owners love once they see the mechanism. Alex cautioned that capitalism doesn't have to be labor exploitation if equity alignment is built in. This protocol is presented as a way to prevent the coming organizational singularity from becoming socially explosive.

Alex described Opus 4.7 as ‘moderately interesting’ but not mythically so, likely a distillation of a larger model. Dave tested it in agentic scenarios; it actively suggested larger budgets and installed a new terminal without prior setup, something 4.6 could not do. The model now accepts images 3× larger, vital for corporate diagram scanning, yet still cannot generate images—a deliberate economic trade-off. The most radical change is the removal of all hyperparameter dials. The documentation bluntly says: ‘Temperature equals zero never was deterministic in the first place’ and instructs users to prompt everything. This forces a new paradigm where natural language is the only lever, raising stakes for prompt design but reducing complexity. The model also showed slightly reduced misaligned behaviors, and Anthropic’s separate paper showed that a weaker model can align a stronger one, which Alex interpreted as a proxy for humans supervising superintelligence.

Peter detailed the company's decline from a $4B valuation to $150M in sales, then a fire sale of its brand. When the AI pivot was announced, the stock skyrocketed. Dave supported the move if the company genuinely puts capital into the AI funnel, e.g., cutting a revenue share with a data center, noting that many successful companies pivoted radically (Nokia tires to phones). Salem warned that narrative leverage must not outpace business model leverage. Alex noted that capital is chasing AI stories faster than operating reality, and that careful execution will determine whether it's pets.com or a real transformation. The discussion underscored market irrationality but also the real opportunity for incumbents to reinvent themselves.

Dave debunked water usage fears, explaining data centers recirculate water in a closed loop unlike chip fabs. He proposed simple fixes: force data centers to generate their own power or pay differential rates to lower consumer bills. Salem argued that constriction drives innovation, as with China's domestic chip development. Alex saw an upside: if ground-based bans persist, it accelerates the deployment of orbital compute (SSO Dyson swarm), which he believes will put the US in the lead. New Hampshire bucked the trend by passing an AI right to compute. The segment exposed how misinformation fuels NIMBYism, creating a political risk that could redirect entire infrastructure buildouts to space.