Filtered for formation
09.14, Friday 15 Sep 2023 Link to this post
How fast does time move in fiction?
I started to wonder how much time passes, on average, across a page of a novel. Literary-critical tradition suggested that there had been a pretty stable balance between “scene” (minute-by-minute description) and “summary” (which may cover weeks or years) until modernists started to leave out the summary and make every page breathlessly immediate. But when I sat down with two graduate students (Sabrina Lee and Jessica Mercado) to manually characterize a thousand passages from fiction, we found instead a long trend. The average length of time represented in 250 words of fiction had been getting steadily shorter since the early eighteenth century.
average length of time described in 250 words of narration was:
- some days, in books written in the 1720s e.g. Gulliver’s Travels
- just under an hour, in the 1990s
The original research was done by hand (well, by grad students, same same). That article is about whether AI can be effective as a tool to make the same estimations…
…which makes me wonder, not having any grad students of my own, what other hidden numbers can be extracted from prose by GPT-4?
I read the other day about the origin of flint, which is wild.
The geology of where I grew up was chalk. Meaning that, as previously discussed, my mental picture of a “stream” is actually a chalk stream, and it turns out this is globally peculiar.
Chalk was formed deep under the ancient oceans of the Earth, from vast quantities of compressed microscopic plankton.
There was also a lot of flint. Using flint for tools in the palaeolithic made sense to me because, well, flint litters the landscape. Not true! It is rare! But, for me, as a kid, you wonder at the colours of flint (and try to smack things with it), draw with chalk, and make things out of the clay you dig up.
Anyway, flint! It never really occurred to me to ask why chalk and flint are co-present.
Apparently the formation of flint was a mystery until the 1980s.
It’s also oceans, it turns out.
Now get this:
Flint was formed in soft, limy mud on the floor of the Chalk Sea some 80 million years ago. It is made of quartz, or silica, which came from the skeletons of tiny sponges that lived in this tropical sea. Their skeletons were dissolved in the seawater and therefore the mud on the sea floor contained silica in small amounts. … Some flint beds can be traced for hundreds of kilometres.
The bizarre shapes of flint nodules, with spiky protrusions and holes, are thought to be due to flint replacing the chalk in the burrows of marine animals such as arthropods that were living beneath the Chalk Sea floor, and it was this connection with burrows that proved to be the key to how flint was formed. The process of flint formation was originally the subject of much argument and was only finally worked out in the 1980s when it was established that flint formed preferentially in burrows due to the presence of decaying organic matter, and also at the ‘redox boundary’, below which anaerobic, sulphate-reducing bacteria predominate. The shape of a flint nodule is therefore often the shape of an animal’s burrow, with the surface often showing the burrowed fabric of the chalk it has replaced.
Not only does flint represent 80 million year old sea sponges, but the shape of the rock is a cast of an arthropod burrow?
And then it was the key material for tool-making by humans for about 3 million years, driving trade routes, and allowing hunting, fabrication, and all the rest?
Deep time vertigo.
Amelia Wattenberger, AI research engineer, has recently prototype a writing app thathighlights sentences according to how abstract or concrete they are.
Wattenberger quotes Robin Sloan, friend of this parish, about the ladder of abstraction:
Good writers move up and down a ladder of language. At the bottom are bloody knives and rosary beads, wedding rings and baseball cards. At the top are words that reach for a higher meaning, words like freedom and literacy. Beware of the middle, the rungs of the ladder where bureaucracy and technocracy lurk.
So… she can reveal that? Astounding.
Read: Getting creative with embeddings (2023) – and check out the screenshots.
This works because each sentence can be placed, using large language models, in “embedding space”, a 1,000-dimensional space where two sentences that mean roughly the same thing will be close together, even if they use different words. But also all the abstract sentences will be closer together versus the concrete ones, and so on.
I realised the other day that this is how I read long articles online:
- read the headline
- read the penultimate para
- scroll wildly up and down reading odd words
- if it’s good, read backwards for a couple sections
- if it’s still good, read forwards from ~1/4 of the way in
- ignore the beginning/end
I’m not saying this is right. I’m just saying that it turns out that this is what I do. It’s probably a symptom of some attentional dysfunction.
Secretly I think it’s because a lot of long-form is filler (including my own). I need to squeeze the fruit before taking it home.
But I wonder whether there are hidden patterns in the essays that I do like, that Wattenberger’s work could reveal:
- speed of ideas: per-sentence velocity through embedding space
- volatility in the abstract/concrete mix – could we Fourier transform Wattenberger’s visualisation and see what frequency mix I like?
- a new scale that shows semantic opposition/agreement vs my own corpus of notes.
Can’t wait to try some of these experiments myself. We are so early with this new technology. We’re imagination bottlenecked. There’s low-hanging fruit for the next decade.
From Greg Egan’s 1997 novel Diaspora, this is the first chapter, about the birth of a machine intelligence - not quite an AI, but a human hosted on a computational substrate - from its own point of view.
Like a baby learning how to use its eyes and hands, and working towards establishing self-identity.
Don’t skip bits like I would, it’s well worth reading properly.
Orphanogenesis by Greg Egan.