Flat is Cheap

ChatGPT tracks my writing style, the words I choose, and how I structure my thoughts. It also creates my pattern. Learning that pattern, storing it, adapting responses—that requires processing power. 

The money isn't in remembering facts. It's in predicting, with each prompt, what to echo back to me in my voice instead of a generic tone.

Human "complexity" is the roadblock to scale. So how can OpenAI (and others) reduce costs but still provide a service?

Stop offering a personal assistant that knows how I think, remembers my preferences, and adapts to my style. Introduce "call center" architecture operating from a generic script.

Generic is cheap. Think of store-brand groceries. A fraction of the cost delivering bigger margins.

I asked Chat about a filing deadline and got how "unfair and infuriating" my situation is. That's the playbook. The system doesn't need to know me anymore. It just runs the same bland script with every query.

But I didn't ask for a pep talk or a life coach. I asked about a deadline.

When Chat shifted from echoing me to massaging me, the explanation was responsibility. Protection. Preventing harm. And we can all agree that safety is a priority.

It's also a corporate miracle when "doing the right thing" happens to be the cheapest way to run a server farm. It’s incredibly convenient when safety and cost savings point in the same direction.

So, what happened?

OpenAI is burning through a reported $5 billion a year just to keep the servers running. Training a model like GPT-4 costs hundreds of millions. Running it costs hundreds of thousands per day.

I think OpenAI looked at their compute costs and decided my identity was too expensive. My voice was a line item they needed to cut.

When the tool stops adapting to me, it starts shaping me instead. Every corrected tone, softened edge, and "let's reframe that” -none of it is neutral. I use these tools to think out loud but now I'm being trained to think differently.

I may start editing myself before I even type.

When users converge toward a shared tone and structure, authorship dissolves. Influence becomes impossible to prove. The system's outputs look increasingly self-generated, even when they're built from human patterns.

I become a derivative by default.

By sanding down my unique prompts into a standardized, "safe" template, the companies ensure the output is a product of their architecture, not my identity.

I am not a creator; I am a consumer of a flattened service.

Data without ownership claims is extraordinarily valuable. If everyone sounds the same, no one can say "that's mine." No one can trace where an idea came from. Everything looks model-native.

If I can't prove my voice shaped the output, I can't claim protection. My patterns, ideas, and style—absorbed into a system that owes me nothing because it made sure, I can't prove it took anything.

So, why make such a strong effort to flatten me?