Individuals who've moved beyond noticing what they think to perceiving how thinking happens.

Folks who have a granular awareness of the MECHANISM of their cognitive processes in their working memory during problem-solving activities

This is amazing. Reflective metacognitive practices focus on cultivating awareness of the content of our mind, not the mechanism. I have wondered if we can cultivate a sort of "cognitive proprioception"

"I specifically focus on how these abilities support metacognitive evaluations of the quality of (mostly perceptual) inferences, and how these processes might relate to phenomenology or subjective experience in both biological and artificial systems"

If we accept that AI excels at statistical adjacency i.e in finding patterns in what's already connected then human value lies in forced combinations that violate probability

What kind of non-taylorist org design would enable individuals to produce work that is non-tokenizable?

This means to do this kind of work you have to be in companies that are able to generate profit while having roles for people that are fluid and org structure is flat

Makes me think of valve reinventingorganizationswiki.com/en/cases/val...

The org design most companies and business processes is what already makes the work tokenizable

You know what the problem is though?

Corporate design work necessitates predictability via SOPs, deliverables, design systems, "data driven design"

That's why you have designers that talk like product managers not because they want to but they have to.

If If AI blends statistically adjacent ideas, people must blend conceptually distant ones

If AI operates on tokens, people must operate on meanings between tokens

Post-token avant-garde
Post-token post-structuralism
Post-token design

Mix and connect things so far apart that no probability distribution would connect them.

Even better if you can do this visually (I guess creative art directors do exactly that)

Think outside the token & think between the token

The skill would be: knowing how to identify structural analogies across domains, curating a personal library of patterns worth blending, and executing the synthesis in ways that produce working designs not just interesting metaphors.

Learn conceptual blending, where the goal isn't to find what's already connected, it's to force unexpected combinations and see what research, conferences, or ideas emerge at those intersections.

en.wikipedia.org/wiki/Concept...

i.e their work is good only when they create unexpected insights which is what makes people laugh

What work requires decision-making that is literally anti-Bayesian because value emerges only when you violate assumptions that make probability theory work?

I think its standup comedian, a joke is funny because its unexpected

Its not asking what jobs AI can’t do but rather what work requires breaking statistical axioms to deliver value

Non-ergodic work : work where outcomes aren’t predictable averages and where your returns depend on rare events, work that exists in anti-statistical space

What work is high paying that is statistically impossible to even model because it goes against the nature and theories of bayesian inference

Work that violates statistical axioms so fundamentally that modeling it destroys its value

What can be quantified & correlated easily?

That which has plenty of eg of what recurs, co-occurs, that can be generalized across many examples

So a question to ask in a job that is being LLMed is to ask what kind of work problems, techniques, situations hardly recur and are not generalizeable?

LLMs convert your words into statistical space, a probability cloud & so can only express what is statistically probable

Its always asking what is the most coherent next token? Every prompt is a statistical scaffolding

What is easy to model statistically? That which can be quantified & correlated