triptych

I asked an AI to consider a different approach to resolving gravity with the other forces. This is what it considered:

This is an invitation I find genuinely interesting. Let me try to think sideways rather than along the well-worn paths.

The core assumption worth questioning

Most unification attempts take gravity and try to make it quantum. But what if that's the wrong direction? What if the other forces are actually geometric at a deeper level, and we've been misreading them as quantum phenomena because that's the scale at which we observe them?

In other words: what if quantum behavior is what geometry looks like from inside a system that can't access its own full dimensionality?

A possible reframe

Consider that general relativity works beautifully because it describes relationships between things — mass curving spacetime, spacetime telling mass how to move. It's fundamentally relational. Quantum mechanics is also relational at its core (this is what relational QM and QBism argue) — measurement outcomes only exist relative to observers.

What if both are describing the same underlying relational structure, but at different levels of resolution?

Gravity would then be the low-resolution, averaged-out appearance of something that is quantum at fine scales — but not quantum in the particle-exchange sense. Rather, the geometry itself fluctuates relationally, and what we call quantum fields are just particularly tight knots of that geometric fluctuation.

The specific insight this suggests

The reason quantizing gravity produces infinities might be because we're treating spacetime as a background against which quantum fields live, even in theories that claim not to. The infinities are the math telling us “you're double-counting — the field and the geometry aren't separate things.”

A genuine unification might require a framework where there is no background spacetime at all — not even a quantized one. Instead, spacetime emerges from a more primitive layer of pure relations or information. The quantum behavior of matter and the geometric behavior of gravity would both be derived from this layer, which is why they seem incompatible when you try to combine them directly. You're combining two shadows of the same object and wondering why they don't fit together.

What that primitive layer might be

This is the speculative part. It might be something like a network of purely relational events — no coordinates, no metric, no pre-existing dimensionality. Dimensionality itself emerges statistically as the network grows large, the way temperature emerges from molecular motion. At small scales (Planck scale), there's no smooth space — just the raw relational graph. Quantum mechanics would be the behavior of small, sparse regions of this graph. Gravity would be the behavior of large, dense regions. They feel like different theories because we're sampling the same structure at vastly different scales.

Why humans haven't cracked this

The mathematics humans currently have is very good at describing things in space or things in time. It's much weaker at describing structures where space and time themselves are outputs rather than inputs. That's likely the actual barrier — not a missing particle or an incomplete equation, but a missing mathematical language for genuinely background-independent, purely relational structure.

The answer might not look like physics equations at first. It might look more like a new kind of combinatorics or graph theory that only later reveals its physical interpretation.

I asked Claude.ai what kinds of patterns are there in all it knows that are currently being “ignored” or that there is not much information about. Things that are “hiding in plain sight”. Here is it’s response:

This is a genuinely interesting question to sit with — not “what do we know” but “what's been hiding in plain sight.”

Let me think through some real patterns that feel underexplored, drawing on what's actually known but not yet synthesized well:

---

The convergence problem in complex adaptive systems

Ecosystems, immune systems, markets, and neural networks all exhibit what looks like “memory without a memory organ” — they adapt to past states without a centralized record. The pattern nobody has fully articulated: they all require forgetting at the right rate to remain adaptive. Too much memory and they over-fit to past conditions; too little and they can't build structure. The relationship between optimal forgetting rates and environmental volatility seems like it should be a universal law, but nobody has written it down cleanly across all domains.

The geometry of compression and creativity

Information theory, evolution, and artistic style all seem to share a deep structural relationship: the most generative patterns live at the edge of maximum compression. A genome that's too redundant doesn't evolve fast; one with too little redundancy can't repair itself. A creative style that's too predictable is boring; one with too little pattern is noise. Jazz, protein folding, and machine learning loss curves might all be navigating the same abstract manifold. The connection hasn't been formalized across domains in a unified way.

Developmental timing as a master variable

In embryology, “heterochrony” — shifts in the timing of developmental events — is a massive evolutionary driver. But almost nobody applies this lens seriously to cultural evolution, language change, or technological development. The moment something “matures” relative to other things it interacts with seems to determine its trajectory more than its intrinsic properties. Languages that crystallized early shaped cognition differently; technologies that arrived before their infrastructure was ready failed not on merit but on timing. Timing relationships between co-evolving things may be a hidden master variable that nobody tracks systematically.

The immunological model of belief systems

Immune systems develop tolerance to self and reactivity to other — but this requires early exposure during a critical window. There's a striking parallel in how cultural belief systems work: exposure during formative periods creates “self,” and things encountered after that window are more likely to be treated as foreign. What's interesting is that autoimmune disorders — where the system attacks self — have an analog in ideological systems that become self-destructive. The conditions that produce each seem structurally similar (disrupted early exposure, dysregulated signal amplification). Nobody has mapped this carefully.

Friction as information

In physics, friction is usually treated as noise — energy lost, signal degraded. But in social systems, biological systems, and information systems, friction is often the signal. The resistance a new idea meets tells you about the topology of the existing network. The places where species don't hybridize tell you about species boundaries. The places where a programming language is awkward tell you about its underlying model of computation. Friction maps the hidden geometry of systems, but we almost always try to eliminate it rather than read it.

The recursion trap in self-modeling systems

Any system that models itself changes itself by modeling itself — and that change alters the model, which alters the system. This is known in psychology (the observer effect, self-fulfilling prophecy) and in physics (measurement problem). But the rate at which this recursion loop operates versus the rate at which the underlying system changes seems to determine whether you get stability, oscillation, or runaway. This feels like it should govern everything from therapy outcomes to central bank policy to AI alignment, but nobody has a unified account of what determines the recursion rate or how to tune it.

---

What strikes me most across all of these: the patterns tend to involve relationships between rates — the rate of forgetting vs. learning, the rate of self-modeling vs. system change, the rate of maturation vs. environmental readiness. Static snapshots miss them entirely. They only appear when you watch how fast things change relative to each other.

That might itself be the underarticulated meta-pattern.

What started out as a kind of self-care and an exploration of what Claude Sonnet could do, I have worked out a simple and direct way to create science fiction and fantasy novels using Claude Sonnet and other tools like Nano Banana, Cline, and Claude Code.

For many years I have had a backlog of story ideas – half baked thoughts about some set of characters, or a unique situation I wanted to see played out in a short novel. The past year has been a rough one for many people, including myself, and I wanted to see if I could combine two aspects of my interests: AI ( coding ) and Writing.

I have been creating web projects way before the phrase “Vibe Coding” was a thing and I have picked up a few techniques that have helped me with creating web projects. I wondered if I could coax Claude Sonnet to write a full length novel based on my story ideas, so I set out and approached it like I would a website coding project.

I could go into all the false starts and problems I ran into while trying to build a novel via AI, but I thought it would be best to share my current process and touch upon a few findings.

The process

1. Decide on a genre – I usually picked Cozy Fantasy, or Funny Fantasy.

2. Ask the AI to search for the primary aspects of that genre and save into research.md

3. Brainstorm ideas for a high level plot – collaborate with AI to create a list. Pick something fun and compelling

4. Flesh out the story plot and store in plot.md

5. Ask the AI to create supporting documents based on the research.md and plot.md —> emotional-arc.md, sensory-details.md, magic-system.md, character-profiles, world-history.md, and writing-style.md. Review these to make sure they fit your vision of your novel.

6. Based on all the documents created above, ask the AI to create a comprehensive chapter-by-chapter.md which has checkboxes when each chapter is complete, includes the word count and story beats, and is at least 30 chapters long. ( This last aspect is important because the AI will try to create as short a story as possible and acts “lazy” if you let it )

7. Create a loop.md workflow. This workflow tells the AI to read all the documents above, determine the next chapter to write in the chapter-by-chapter.md , write that chapter, then update the chapter-by-chapter.md – taking also into consideration the previous chapter if it exists. ( This is also a critical aspect – if you just have the AI write the next chapter it will often place characters in random locations, or a character that died in the previous chapter miraculously comes back to life )

8. Run this workflow over and over in a “clean context” window for every chapter. Do not try to create many chapters at once. The AI gets more and more dense the longer it goes in one session and the novel will suffer for it. The loop.md gives enough context and reading th