The LLM Gamble

LLMs behave like slot machines — nondeterministic outputs create a dopamine-driven feedback loop where users remember wins and forget failures. The financial model compounds this problem: users pay for both input and output tokens regardless of quality, output length is unpredictable, and subscription plans obscure true usage limits. Since inference costs money whether the response is useful or not, current pricing (likely below cost) is unsustainable. The author argues that pay-per-use models are hard to sell when users can't predict costs or outcomes, posing a long-term viability threat to the generative AI industry.