← back

On Letting Go

The universal human impulse is one of variable reduction - or, put less ostentatiously, understanding. The world is filled with unbearably many sights, smells, and sounds that beg to be compressed. In our expeditions to comprehend our short, insignificant lives, humans are generally driven to discover “ground rules” for our surroundings and world as a whole. This is, after all, the crux of mathematics, physics, and philosophy - developing theories that can be generalized to various situations.

I won’t deny other impulses: pleasure, safety, growth, love. But these, I believe, are downstream of a craving to understand. We have uncountable means of doing so - computer simulations, experiments, the Standard Model, the paragraphs in your Notes app. And the prevalence and continual acceleration of AI/AGI progress have only amplified the tools at our disposal. But there is one medium that I believe deserves greater attention, especially on the road to AI 2027: art. Art, the soul’s interpretation of a changing world. So, having made blanket statements about humanity, I’ll offer a few comments on deriving meaning from great art in an age of long-horizon agents, recursive self-improvement, and other buzzwords.

Looking Back

But why care about art? my younger self would’ve asked. Why not build? Explore? Travel? Because art - no matter the medium - is a vessel for understanding our worlds. There it is again…understanding.

My Grand Unified Theory of *great* art: concise, surprising, and relevant. Concise, as in maximizing insights per word/brushstroke/thing to express the author’s wisdom, not the two-page limit for your consulting club case study. Surprising, because it activates dopamine receptors, and relevant because people are egocentric. Intelligence becoming abundant begs the question: will the (human) ability to transmute our experiences into great art turn sacred or scarce?

At this point, the especially picky reader may observe a contradiction. Say you trust my thesis about humans seeking understanding. Then why do we find surprising art appealing…after all, we want to understand the world, not continually be assaulted with new phenomena! The resolution, I think, requires us to zoom out. Surprises, whether you like them or not, are unavoidable - but not all are created equal. Exposing ourselves to surprises early lessens (or even eliminates) the outsized impact of later ones. Discovering a partner’s infidelity in year one is preferable to discovering it on the security footage in year twenty. Or more tamely, a foundation model of the cell shifts surprises earlier (e.g., the drug fails now instead of in the Phase III trial). I realize I’m stretching the definition of ‘surprise’ - but I think my point still stands that surprise has time value. Art, then, is the lowest-stakes mode of surprise, as you are inoculated against new/upsetting ideas and experiences in the safety of a sandbox. It may be useful to think of your brain as a compressor; something is beautiful in proportion to how well it compresses under your current model (example: symmetric faces), and interesting in proportion to how fast your compression of it is improving (e.g., solving puzzles).

So art is where you turn to be purposefully surprised. Anyway, at risk of becoming an armchair philosopher, I’ll move to more tangible ideas.

Looking Forward

In “The Inner Ring,” C.S. Lewis urges readers to let go of the incessant desire to be inside vs. out - there is always a more exclusive group. As frontier models not-so-slowly creep forward in capability, much expert brainpower has been devoted to imagining the future of art in an age of superintelligence. LLMs may not be stochastic parrots, but can they craft truly novel art? Yet what if in our fixation with crafting beauty, we’ve forgotten to step back?

So here’s a paradigm: frontier models themselves are art! We’ve established that exceptional art is concise, surprising, and relevant. Relevance is apparent, but surprising? I’d call myself duly surprised when learning about how GLM-5.3 reasons about how its current actions could become training data and impact its future self, or when reading about the agent message board in the OpenAI-Hugging Face fiasco. In the broader sense, I’d argue that frontier models are becoming increasingly surprising in the form of scientific breakthroughs, misaligned behavior, and mathematical superintelligence. Concision may be argued as well: with our insight per unit definition, a frontier model is (one of) the most compressed objects humans have ever built - a few terabytes of weights standing in for the internet. So perhaps our attention should be turned toward deciphering the anatomy of these trillion-parameter pseudo-brains.

In coming years, I’m confident that discourse around AI and art of all forms will grow exponentially and loudly. But in the process, I presumptuously urge you to fall back on your definition of art and stop climbing the conventional ladder of proving status. Oftentimes, there is too little at the top. Instead, somewhere along the way, consider opening your palms and letting go. You never know - you might just land on a staircase to someplace new.