Artists: "My job will be gone in xx months!"
AI labs: "We're automating everything tomorrow!"
Meanwhile, there's an invisible layer no one's talking about.
Tyler Cowen just asked Brendan Foodie from Mercor a question that cracked open my thinking:
"Should we be training AI on contemporary taste?"
Should we even be using people from NOW?
What if we pulled from peak eras instead?
Not Pauline Kael (she's dead).
Not the peak Rolling Stone writers from the 80s (imagine Hunter S. Thompson clicking A or B? Imagine the writing ChatGPT COULD have).
Not Cahiers du Cinéma critics who defined modern film theory.
Can we get Scorses in an RL environment, please? (while we still can)
What we have instead is an opaque workforce. Tens of thousands of people clicking buttons for cash.
Tyler's point: Maybe we're not at a cultural peak right now. And their taste is transiently good.
Are today's film and TV at their peak? Today's music? Today's writing? If not, we're encoding a somewhat mediocre moment into the foundations of our future thoughts.
Here's why this matters:
Brendan explained that even in law - a very prescriptive field - models have a fundamental issue:
"There's a lot of areas in law where the right way of approaching something is not written down or codified. It exists more in the heads of experts. And I think it's those domains where there's a lot of taste that isn't well-documented, that the models will struggle immensely with."
If this is true for LAW, what does it mean for creative work?
The judgment that makes something "good" lives in our heads. Each head has its own weights. Its own calibration. It's own moment in culture. And right now, we're encoding that judgment through whoever happened to be available for RLHF work in 2025.
Full transparency, I've done some work in RL for Remotasks back in the early days.
This is The Taste Trap:
→ 2025 contractors click "prefer A" thousands of times
→ That becomes training data
→ AI learns what "good" means from today's contractors
→ Every creative tool builds on that foundation
→ For how long? We don't even know.
We're enshrining people. RANDOM PEOPLE.
It's a nuanced but important point.
Brendan suggests we're heading toward a world where everything becomes a "staged RL environment" - constantly teaching and retraining agents.
In this vision of the future we will all just work for Mercor.
The interesting question is: if we start our foundations from a mediocre 2025 taste, aren't we just building on an uneven, spiky foundation?
Even if it's not permanent, the starting point matters.
The implications are already here:
This is why AI writing has that distinctive taste you can instantly spot.
This is why AI has spiky intelligence.
This is why people don't trust AI to write with a real POV.
This is why everything feels like it came from the same aesthetic.
We're not asking the right questions:
Is 2025 the right era to enshrine as our foundation?
Are these the right people?
Is this moment actually the one worth enshrining?
Are we going all become RL inputs?
The conversation on X: Artists vs AI labs.
How can we make sure creative mediocrity doesn't become permanent?