Back on my theme of trying to understand why LLMs have taken off:

Puzzle piece 1: LLMs are, in Frankfurt's sense of the term, bullshit machines. To him, a core component of bullshit is "indifference to how things really are".

Puzzle piece 2: Managerialism, a dominant business philosophy, holds that managers are universal. They can manage anything without regard to the topic. Without truly learning "how things really are".

Puzzle piece 3: On average, the larger the company the higher the density of bullshit. CEO statements, managerial presentations up and down, customer comms, etc. Related is the density of office politics, careerism, etc.

So putting this together, it seems like our corporate hierarchies have a hard time understanding the weak point of LLMs because standard corporate culture has the same weak point: valuing many other things over the truth.

This, via @davidgerard, is a #bluesky gobsmacker.

First there's the breathtaking confidence that this this exciting AI thing will solve the problem. As if Twitter didn't have literally hundreds of ML engineers working on this for years. YEARS.

Then there's the notion that you could possibly create AI magic without doing enough of the manual work to really understand it. Nope!

Third, there's apparently a belief that a magic technical solution exists to social problems that are so complex that they're demoralizing and hard to deal with.

Fourth, I see no recognition that they should have had at least some starter solutions before letting users on at all.

Lastly, there's the lack of recognition that anti-abuse work, hard though it is, is not nearly as hard as dealing with the abuse.