Why I'm still bearish on LLMs after Navier-Stokes

Frontier AI might be impressive, but exorbitant costs mean early-stage founders should think twice before building core products on hyped-up LLMs.

A recent essay titled "Why I'm still bearish on LLMs after Navier-Stokes" sparked a heated discussion on Hacker News, drawing hundreds of points and comments. The core takeaway resonated with a growing crowd of pragmatic builders: while current LLM capabilities are genuinely impressive, frontier models remain ludicrously overpriced for the value they actually deliver.

For early-stage founders, this gap between raw capability and unit economics is where startups live or die. Burning your pre-seed runway on eye-watering API costs just to plug into the newest frontier model isn't a sustainable business model. If your product's margins rely on expensive, heavy infrastructure, you are effectively subsidizing your users at the expense of your own survival.

For small backers investing from £500, this commentary serves as a crucial filter when evaluating pitch decks. Don't be dazzled by founders who simply wrap an expensive LLM and call it innovation. The teams that build lasting value will be the ones focused on disciplined unit economics, pragmatic model choices, and real cash flow over raw technical hype.