That seems to be exactly her point, no?

Imagine an interface that reveals the engagement mechanism by, say, having an additional iframe. In this iframe an LLM clicks through its own set of recommendations picked to minimize negative emotions at the expense of engagement.

After a few days you're clearly going to notice the LLM spending less time than you clicking on and consuming content. At the same time, you'll also notice its choices are part of what seems to you a more pleasurable experience than you're having in your own iframe.

Social media companies deny you the ability to inspect, understand, and remix how their recommendation algos work. They deny you the ability to remix an interface that does what I describe.

In short, your quote surely applies to social media companies, but I don't know if this is what you originally meant.