Ronan Farrow's New Yorker feature on Sam Altman, discussed this week on Nilay Patel's Decoder podcast, centers on what Farrow calls Altman's "unconstrained" relationship with the truth. The timing is either ironic or perfectly chosen: the same week, researchers published a paper finding that large language models exhibit numerical instability and chaotic unpredictability that makes their outputs fundamentally unreliable in ways that compound across uses. The CEO and the product share a structural characteristic: both present confidence as competence.
Quantifying Untruth in AI Systems
The arXiv paper by Islam, Villarreal, Nishino, Salman, and Liu on numerical instability in LLMs found that small perturbations in inputs produce large, unpredictable divergences in outputs, a property the researchers explicitly compare to chaos theory. This is not hallucination in the colloquial sense. It is something more fundamental: the model's outputs cannot be reliably reproduced or verified. Scale this up to the agentic workflows that SciFi's autonomous scientific AI paper describes, and you have a trust infrastructure problem. You are being asked to trust outputs from systems whose architects openly acknowledge they cannot fully predict or audit. Farrow's critique of Altman lands differently in this context: the epistemological culture of the organization may be continuous with the epistemological properties of the product.
Attention, Accountability, and Eric Swalwell
The trust economy extends beyond OpenAI. The New Yorker's analysis of Eric Swalwell's political collapse attributes it to the attention economy's logic: performance over substance, visibility over reliability. The same dynamic that produces viral politicians produces viral AI products. For founders navigating investor trust in an environment saturated with AI hype, TurboFund's guide on targeted fundraising argues that signal quality beats broadcast volume every time, which is basically the anti-Altman pitch. Cory Doctorow, writing in The New Yorker this week, frames the billionaire trust problem as structural, not personal. It is not that Altman is uniquely dishonest. It is that systems which concentrate power reward a particular relationship with truth.