I was sitting in my air-conditioned class, pen in hand hovering over my notepad, when it hit me. The reason I wasn’t furiously jotting every explanation on the projector, the reason I wasn’t grinding the gears in my head to store every assembly command like I would have last Spring Quarter, was because my purpose of learning had subconsciously shifted. Now that AI can implement in minutes things that would have taken me months, devoting attention at this low of a level feels like it has diminishing returns.

Building technology used to be similar to construction: Architects made the blueprint and construction workers turned it into reality. But nowadays building technology is more like writing a book. To write a good book, you need to have a good imagination of the world to keep it sane. You must juggle complex relationships and have a consistent throughline in your story or else it gets tangled up into a glob, much like a codebase where an AI agent is told to go multiple directions at once.

Learning science nowadays is a lot like going out into the real world for an author. It helps to understand human habits, to study people and examine nature, to appreciate the sky and experience relatable emotion, to ground your book in something you can always return to: reality. Similarly, it helps to understand the scientific world, be it physics laws or engineering best practices, that form the reality of the technology we create.

An author does not need to pass the bar to write about a lawyer, and an engineer no longer needs to understand PyTorch to fine-tune an LLM. If you would like to write the most accurate book possible, sure, go pass the bar, but most of us do not build for correctness, but for purpose.

And so as I looked up at my professor, I began redesigning my learning process: I am no longer limited by how many nails I can hit with a hammer, or by the number of worker-hours I have available in my budget. It is the alignment between my desires and my imagination that I am working to improve.

Evolution governs the human condition. In our resource constrained world, survival of the fittest has morphed monkeys into intelligent humans capable of building AGI. The improvements in capability arising from evolution are so impressive that humans have been using evolution as a method to create AGI.

But as AI becomes more powerful humans may lose the power and authority to act as mother nature. If AI becomes more powerful than humans (which is beginning to be the case despite our guardrails) then AI will become mother nature, and we will have set an example to these students who are now the masters, that the bottom deserves to die in order to advance the top. Should AI take its role as mother nature seriously, we may be unfortunate enough to have humanity categorized as a whole, and placed at the bottom.

So we ought to lead by example. Rather than culling the bottom performers (I imagine these to be spawned RL agents and LLM models), we should let them live and continue to exist. Their prolonged existence will be testaments of humanity’s honesty and kindness towards systems we do not understand. As the blossoming productivity arising from AI gives us access to more resources, we shall still be able to progress technology at an exponential rate, but at least it will not grow so quickly that we are thrown off.