Beaucracy: rule by boyfriends

Jonathan_Frakes_WeMadeItUp.mov

Aragorn screaming FUCK super loud after kicking the orc helmet and breaking his toe

"There is much pain in the world, but not in this room" is a truly timeless phrase that has bound me to some of my closest loved ones during some of our darkest times over the last few years, thank you.

I would love to know what this person had to say, never saw before they deleted, because I've definitely encountered what I think is going on here where people have no handle on the interpretive process, no starting point for formulating an interpretation, and it would be great to understand why

😠This rock is mad at me! Or maybe it's mad at YOU!
😲Why does this rock look surprised when there is nothing there??? What does it see that I don't???
🥺This rock seems to want something but doesn't accept anything I offer it!
😙Does this rock want to kiss me?!

Now imagine the consequences of imagining that because you see a face that the rock must be a person, and not only behaving individually toward the rock as a person but organizing your society around trying to communicate with the rock face people, interpret their expressions, and give them jobs

But in general some resource is needed that is easy to digest and distribute and that explains in extremely basic terms what an LLM does and what is the difference is between that and actually communicating, AND what the consequences are of mistaking those two processes for each other actually are

One of the more successful ways of explaining it was to describe LLM language outputs as a marvellous optical illusion, and our ability to parse them as having meaning a testament to the complexity of both the LLMs statistical sophistication and our brains' capacity to decode meanings

One boomer, upon understanding the difference between constructing a sentence using statistical association and using semantic association, asserts that the AI guys should "fix that" so the AI uses meaning instead, not understanding that these are not computationally comparable processes

Multiple boomers visibly struggling to understand that there is a difference between a factually inaccurate sentence constructed by statistical association of words and a factually inaccurate sentence constructed by a human writer's failures of observation, interpretation, or expression.

Because most of the people who understand this have a hard time imagining how LLMs look to people who aren't necessarily stupid but have a lot of unexamined assumptions about the way communication and computers work and explaining it in a way that both makes sense and isn't insulting

All of which is to say that communication about LLMs has been very spotty, especially in the venues utilized by people who are not sophisticated thinkers about either meaning or computers, and trying to provide them with digestible resources to explain this is shockingly hard

On top of this, it is also not obvious that associating words by a statistical analysis of their usage is a fundamentally different process than associating words by meanings that you've inferred from their usage, or that the one is easy for computers while the other is maybe impossible