A lot of times when dealing with complicated systems, it’s easy to get to a “turtles all the way down” point. What’s interesting with a lot of AI conversations right now is we’re watching ourselves stack the turtles.

I found myself stacking turtles reading this article about the impact of Amazon’s AI summaries on product sales. Basically sellers say these summaries over index on negativity. So I can see the utility in Amazon making these product review summaries, but also how this could really hurt a seller. But because there will be no way to scale any kind of human curation of these summaries, the next inevitable step will be to have an LLM review these summaries for overly negative content. Which then will require this ever spiraling series of LLM review because we are setting up systems that cannot scale to human curation. Turtles are now stacked. It’s supremely bizarre to see this happen in real time.