<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tom Moore on Medium]]></title>
        <description><![CDATA[Stories by Tom Moore on Medium]]></description>
        <link>https://medium.com/@mooreiarty?source=rss-073290cb80a3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*tU7ukYB1-Eym-FdksPrRbA.jpeg</url>
            <title>Stories by Tom Moore on Medium</title>
            <link>https://medium.com/@mooreiarty?source=rss-073290cb80a3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 11:00:26 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mooreiarty/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Russian Blues — A Cat’s Tale in Colorspace]]></title>
            <link>https://medium.com/the-accidental-geometer-the-geometric-whorfian/russian-blues-a-cats-tale-in-colorspace-11af6fa03117?source=rss-073290cb80a3------2</link>
            <guid isPermaLink="false">https://medium.com/p/11af6fa03117</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[philosophy]]></category>
            <dc:creator><![CDATA[Tom Moore]]></dc:creator>
            <pubDate>Sat, 20 Dec 2025 14:16:09 GMT</pubDate>
            <atom:updated>2025-12-20T21:18:01.957Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Russian Blues — A Cat’s Tale in Colorspace</strong></h3><figure><img alt="A stylized illustration of a blue cat in profile, mouth open as if vocalizing, with a network of glowing nodes and connections emerging from its mouth, suggesting speech or information flowing into a semantic network against a cloudy blue and purple background." src="https://cdn-images-1.medium.com/max/1024/1*eKSfunmx6ebuLOjZG6a-bw.jpeg" /></figure><p><em>Note: This essay introduces soon-to-be-published findings, with detailed experimental results and an extended discussion to follow.</em></p><h4>A Talkative Cat, a Talkative Language</h4><p><em>The talkative cat as metaphor: one axis, many signals.</em></p><p>Russian Blues are famously talkative cats. They chirp, trill, and narrate your life like a tiny, sooty sports commentator who is absolutely certain you need continuous updates about the state of the hallway.</p><p>This post is about a different kind of Russian blue, but the parallel is hard to resist.</p><p>We’ve been working on the <strong>geometry of meaning</strong> in multilingual language models, and one of the cleanest case studies lives in colorspace, with a particularly interesting example around <em>blue</em>. English looks at the sky, more or less shrugs, and says “blue.” Russian does not shrug. Russian will happily subdivide your sky into distinct lexical provinces. In a precise sense I’ll define below, Russian is more <em>talkative</em> in blue.</p><h3>What we’re actually doing</h3><p>At a high level, our project asks a straightforward question:</p><blockquote><em>If you learn a big multilingual </em><strong><em>semantic space</em></strong><em> from text, can you </em><strong><em>see cross-linguistic differences</em></strong><em> in how concepts are carved up by looking at the geometry of that space?</em></blockquote><p>Our hypothesis proposes that concepts like colors occupy semantic clusters both in our minds and in large language models, with the two being highly correlated. The clusters themselves consist of words that join together with a particular, almost physical, geometry. That geometry influences how we think and how language models behave.</p><p>Take a modern language model or off-the-shelf multilingual word embeddings. Each word becomes a vector in a high-dimensional space. Words that behave similarly in language end up near each other; words that behave differently spread apart. You don’t tell the model what “similar” means. It discovers a structure that lets it predict words from other words, and the geometry falls out of that learning process.</p><p>If language influences thought even a bit, it should leave fingerprints in that geometry. Not in a strong Whorfian “different worlds” sense, but in a very geometric one: where a language has more <strong>lexical distinctions</strong>, we should see more structure — more clustering, more curvature, sharper boundaries — in the corresponding region of semantic space.</p><p>Color words are a useful testbed because the underlying perceptual space is shared across humans, but languages differ widely in how many basic terms they use and where they draw boundaries. We started with color as a sandbox for what we call <strong>geometric Whorfianism</strong>.</p><figure><img alt="A comparison of semantic vector spaces where English represents blue as one diffuse cluster, while Russian separates blue into two distinct clusters labeled “голубой” (light blue) and “синий” (dark blue)." src="https://cdn-images-1.medium.com/max/1024/1*-9xWAbprjIdJNiqP_iJg3w.jpeg" /></figure><p><em>English “blue” forms a single diffuse region; Russian splits the space.</em></p><h3>The Russian blues</h3><p>In English, we casually use one basic term for a wide band of the spectrum: <em>blue</em>. We can qualify it (light blue, dark blue, navy, teal), but the unmarked word is just <em>blue</em>.</p><p>Russian lexicalizes a split. There is <strong>голубой</strong> (<em>goluboy</em>), roughly “light blue,” and <strong>синий</strong> (<em>siniy</em>), roughly “dark blue.” These are not poetic distinctions; they are basic, everyday terms. Russian speakers treat them more like English treats <em>red</em> versus <em>orange</em> than <em>blue</em> versus <em>navy</em>.</p><p>We asked whether that fact shows up in semantic geometry.</p><p>Using pretrained fastText Common Crawl multilingual embeddings, we isolated the region around color terms and examined cluster structure. The result is clean: the Russian blues split into two distinct clusters. The distance between their centroids is comparable to the distance between distinct basic color categories in English.</p><p>English “blue” forms one stretched blob. Russian forms two. The model wasn’t told these were basic colors. It inferred that from usage patterns. Russian simply does more with that region.</p><p>In the cat metaphor: English mumbles “blue” once and moves on. Russian trills a whole sequence of blue-noises.</p><figure><img alt="A stylized side-profile illustration of a human head with a glowing, network-like brain, containing light and dark blue geometric shapes and a clock icon, suggesting cognition, time, and structured information processing against a technical blueprint-style background." src="https://cdn-images-1.medium.com/max/1024/1*tNunfGX7s1j_P-0G6MYBMw.jpeg" /></figure><figure><img alt="A stylized side-profile illustration of a human head with a glowing, network-like brain, containing light and dark blue geometric shapes and a clock icon, suggesting cognition, time, and structured information processing against a technical blueprint-style background." src="https://cdn-images-1.medium.com/max/1024/1*tNunfGX7s1j_P-0G6MYBMw.jpeg" /></figure><p><em>Behavioral validation connects geometry to cognition.</em></p><h3>The experimental Russian blues result</h3><p>Embedding geometry is suggestive, but it isn’t enough. To claim a Whorfian effect, even in a weak modern sense, you need behavioral validation.</p><p>That’s where the classic study <em>Russian blues reveal effects of language on color discrimination</em> (Winawer et al., 2007) matters. Russian speakers were faster at distinguishing colors across the <em>goluboy/siniy</em> boundary than within it. English speakers showed no such advantage on the same stimuli.</p><p>Crucially, the effect disappeared under verbal interference but not spatial interference. That strongly suggests online language involvement in perceptual decisions.</p><p>This is textbook psychophysics: controlled stimuli, objective tasks, reaction time as an implicit measure. It bridges the gap from “languages differ” to “minds behave differently.”</p><p><strong>[Insert image: russian_blues_concept_space.jpg]</strong><br><em>Languages sculpt geometry rather than inventing reality.</em></p><h3>Languages as manifold sculptors</h3><p>“Language sculpts reality” is an oversimplification. The physical world and our perceptual systems constrain what distinctions are viable. Strong Whorfianism is historically disconfirmed.</p><p>A more accurate picture is this:</p><blockquote><em>The world presents a continuous high-dimensional field. Languages introduce structured partitions that warp the geometry of how we navigate it.</em></blockquote><p>When a language lexicalizes a distinction, it creates stable attractors in usage. Those attractors show up as dense regions in embedding space, and those regions bend the local geometry. It’s a form of cultural symmetry-breaking imposed on a shared perceptual substrate.</p><h3>Why I care</h3><p>This is the kickoff of <strong>Project Umwelt </strong>(<em>more to follow soon</em>).</p><p>The Russian blues are a clean case: same sky, different lexical tiling, measurable geometric differences in learned semantic space, and measurable cognitive differences in humans under controlled conditions.</p><p>The <strong>Geometric Whorfian Hypothesis</strong> proposes that languages impose systematic geometric structure on shared conceptual substrates. These structures affect distances, neighborhoods, and gradients in semantic space, shaping thought and behavior without determining them.</p><p>Understanding those geometries matters not just for cognitive science, but for how we build and align large language models. Instead of flattening linguistic variation into an English-centric space, we may get better results by preserving distinct linguistic manifolds.</p><p>Meanwhile, the cat on the desk is not arguing. She is loudly explaining that the food bowl is 37% less full than it should be.</p><p>Different systems, same pattern: some things in the world are just intrinsically more talkative.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=11af6fa03117" width="1" height="1" alt=""><hr><p><a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian/russian-blues-a-cats-tale-in-colorspace-11af6fa03117">Russian Blues — A Cat’s Tale in Colorspace</a> was originally published in <a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian">emergenesis</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Language of Theseus — How the Lyrics Change But the Song Remains]]></title>
            <link>https://medium.com/the-accidental-geometer-the-geometric-whorfian/the-language-of-theseus-how-the-lyrics-change-but-the-song-remains-17bf6b62319f?source=rss-073290cb80a3------2</link>
            <guid isPermaLink="false">https://medium.com/p/17bf6b62319f</guid>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Tom Moore]]></dc:creator>
            <pubDate>Tue, 25 Nov 2025 23:21:46 GMT</pubDate>
            <atom:updated>2025-11-26T03:53:03.484Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>The Language of Theseus — How the Lyrics Change But the Song Remains</strong></h3><figure><img alt="A sailing ship whose boards are made of words in various stages of aging. Some are brand new, others are old and in need of replacement." src="https://cdn-images-1.medium.com/max/1024/1*xoKktfx0X18ZadVNR_6wyQ.jpeg" /></figure><p>Imagine the Ship of Theseus. You replace one plank, then another, then another, until none of the original wood is left. Is it still the same ship?</p><p>Now swap planks for people.</p><p>Speakers are born into a language, use it for a few decades, and then vanish. New speakers arrive, inheriting the words and habits of those before them. The language community that speaks “English” in 2025 is not the same set of humans that spoke “English” in 1925, and yet we talk as if it’s one continuous thing.</p><p>Cognitive scientist Peter Gärdenfors puts it like this: even if meanings are determined by what’s in the heads of the users, those meanings can stay relatively constant as the users are gradually replaced. A word can seem to survive generations of turnover.</p><p>Call this idea <strong>the language of Theseus</strong>: the apparent continuity of a language even as its human planks are endlessly swapped out.</p><p>But words don’t just survive; they move.</p><h4>When meanings quietly drift</h4><p>We all know, at some level, that words change meaning over time. Historical linguists have catalogued thousands of these shifts, but you don’t need a PhD to feel it.</p><p>Take <strong>“mouse.”</strong><br>For centuries, it’s a small furry animal you don’t want in your kitchen. In the late 20th century, it also becomes an input device you do want on your desk. Same sound, same spelling, new neighborhood of meaning.</p><p>Or <strong>“cloud.”</strong> <br>Once: fluffy thing in the sky, neighbor to “rain” and “storm.” Now: also “the place your files go,” sitting near “data, server, platform” in tech talk.</p><p>Even “<strong>literally</strong>” has wandered. It used to insist “I’m not exaggerating.” Now you can say “I literally died laughing” without anyone calling an ambulance. The word’s job has shifted from precision to emphasis.</p><p>Our intuitions say: these words have drifted. But can we do more than nod and make a few before/after examples? Is there a way to <strong>see</strong> this drift in a more systematic, almost physical way?</p><h4>Words as points in space</h4><figure><img alt="Small, star-shaped semantic networks on a graph. The division down the center splits between 1970 and 2020, showing the transition from a weather-associated word “cloud” to a split definition with two clusters — one for weather-related associations and the other for technological." src="https://cdn-images-1.medium.com/max/1024/1*_F6xriTroSYHomJGYSnX8g.jpeg" /></figure><p>Enter word embeddings.</p><p>Modern language models often represent each word as a list of numbers — its coordinates in a high‑dimensional space. Each word becomes a point in this invisible cloud. Words used in similar contexts end up near each other. “Dog” is close to “cat” and “puppy,” far from “quantum” or “mortgage.”</p><p>You don’t have to love linear algebra to get the basic picture:</p><p>- The model learns a <strong>shape</strong> of language use.<br> <br>- Distances in that shape reflect relationships in meaning.</p><p>Now add one simple twist: <strong>time</strong>.</p><p>Instead of training one model on all English text ever, you train separate models on different time slices: one on texts from 1900–1920, another from 1950–1970, another from 2000–2020. Each model gives you a snapshot of where words lived in meaning‑space during that period.</p><p>These are temporal embeddings — word embeddings indexed by time.</p><p>Line up those snapshots (there are standard mathematical tricks for aligning them), and you can literally watch a word’s position move across decades. Early on, the point for _cloud_ hangs out next to “rain” and “storm.” Later, it begins to drift toward “server,” “storage,” “platform.” The word hasn’t just “changed meaning” in some vague way; its coordinates in this conceptual geometry have migrated.</p><p>The same goes for “mouse,” “literally,” and countless others. For some words, the path is smooth and gradual. For others, you get something like a fork: the word splits into two distinct senses that occupy different regions of the space.</p><p>From this perspective, the language of Theseus is no longer an abstract metaphor. It’s a movie: a point cloud wriggling and stretching as time goes by, words sliding into new neighborhoods while the overall structure of the language persists.</p><h4>The shape of thought: a geometric Whorfian hypothesis</h4><p>There’s a famous idea in linguistics associated with Benjamin Whorf: the languages we speak influence the way we think. If your language makes certain distinctions easy (like many words for snow) and others hard, it gently nudges what you notice, remember, and reason with.</p><p>Now overlay that idea on the embedding picture and you get a geometric twist:</p><ul><li>If meanings live as points in a <strong>geometric space </strong>—</li><li>and that space has a certain <strong>shape</strong> — clusters, directions, gaps –</li><li>then that shape influences which associations feel natural or immediate.</li></ul><p>Call this the <strong>Geometric Whorfian Hypothesis</strong>: the geometry of our conceptual spaces helps structure our thought.</p><p>This is a working hypothesis, not a proven law of cognition. But it’s a useful lens. When a word like “cloud” moves into a new region of the space, neighboring “data” and “server,” it becomes easier to connect those concepts. When “literally” drifts away from careful, logical writing and toward hyperbolic storytelling, it becomes more natural to use it as an intensifier rather than a guarantee of accuracy.</p><p>Temporal embeddings don’t just document change; they suggest how the <strong>affordances for thought</strong> might be changing too. As the manifold — the overall shape of meaning — bends and warps, some mental paths shorten while others get longer.</p><p><strong>Newspeak: hacking the manifold</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KsSPrap6AhviDKIuDpjcew.jpeg" /></figure><p>George Orwell’s <em>1984</em> imagines a language engineered to control thought: <strong>Newspeak</strong>. The idea is simple and terrifying: if you remove or compress words for rebellious, critical, or nuanced ideas, it becomes harder to even formulate those thoughts.</p><p>“Bad” becomes “ungood.” “Excellent” becomes “plusgood.” More extreme criticism might be “doubleplusungood.” Subtle moral vocabulary is flattened into a few graded blobs.</p><p>In embedding terms, Newspeak is like taking a rich, detailed manifold and crushing it:</p><ul><li>Many distinct concepts that once occupied different regions of space are forced into almost the same coordinates.</li><li>Entire directions in the space — along which people used to differentiate ideas like “injustice,” “resistance,” “freedom” — are deleted.</li></ul><p>You still have a language of Theseus in the sense that new generations grow up speaking it, but the ship has been radically rebuilt. Its shape no longer lets you sail to certain conceptual destinations. The geometry has been edited to make some thoughts nearly inexpressible.</p><p>Orwell’s scenario is extreme and fictional, but it sharpens the point: if you can manipulate the geometry of meaning, you can influence what feels thinkable.</p><h4>Seeing the Ship of Theseus</h4><p>Return to Gärdenfors: meanings live in the heads of users, but can remain stable as those users are gradually replaced. With temporal embeddings, we can make that idea less mystical.</p><p>For some words, you’ll see remarkable stability. The point for “triangle” isn’t going anywhere; its neighbors stay like “angle,” “geometry,” “polygon.” The language of Theseus keeps that piece of the ship intact over long stretches.</p><p>For other words, you can trace long, slow drifts or dramatic jumps. The coordinates for “mouse” yank themselves toward “screen” and “click.” The manifold changes not just because individual speakers come and go, but because culture, technology, and power rearrange what goes with what.</p><p>The language of Theseus, in this geometric view, is a structure that both <strong>persists</strong> and <strong>flows</strong>. It persists enough that we can still read Shakespeare, if with footnotes. It flows enough that we routinely underestimate how different our ancestors’ meanings really were.</p><p>Temporal embeddings give us a way to watch the planks being swapped — to <strong>see</strong> the splits, the drifts, and the quiet revolutions in word meaning. And the Geometric Whorfian perspective invites a further question: as our shared meaning‑spaces keep bending, what kinds of thoughts are we making easier for future minds, and which ones are slowly drifting out of reach?</p><p>That’s the puzzle at the heart of the language of Theseus: not just whether it’s the same ship, but where, exactly, it’s sailing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=17bf6b62319f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian/the-language-of-theseus-how-the-lyrics-change-but-the-song-remains-17bf6b62319f">The Language of Theseus — How the Lyrics Change But the Song Remains</a> was originally published in <a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian">emergenesis</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Accidental Geometer — The Geometric Whorfian Hypothesis for LLMs]]></title>
            <link>https://medium.com/the-accidental-geometer-the-geometric-whorfian/the-accidental-geometer-the-geometric-whorfian-hypothesis-for-llms-a2a0aac4180f?source=rss-073290cb80a3------2</link>
            <guid isPermaLink="false">https://medium.com/p/a2a0aac4180f</guid>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[linguistics]]></category>
            <dc:creator><![CDATA[Tom Moore]]></dc:creator>
            <pubDate>Thu, 20 Nov 2025 23:32:05 GMT</pubDate>
            <atom:updated>2025-11-21T00:26:17.087Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>The Accidental Geometer: The Geometric Whorfian Hypothesis for LLMs</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jKC3XgI0zNScbNK0zaZ2fg.jpeg" /></figure><p>If you work on large language models, you already think in vector spaces. Words, sentences, documents, and activations are represented as points in a high-dimensional vector space where cosine similarity stands in for “meaning.” The Geometric Whorfian Hypothesis (GWH) starts from that familiar picture and makes a stronger claim: LLM embedding spaces should be treated as Riemannian, language-conditioned models of human conceptual geometry, not simply Euclidean containers for co-occurrence statistics. Instead of seeing embeddings as a convenient representation of statistical data, we should treat them as our first large-scale, manipulable approximations to how human minds carve up meaning.</p><p>GWH fuses three strands. First, modern **linguistic relativity**: languages and cultures shift how people categorize, infer, and remember. This does not happen by giving them “different worlds” in the classic Whorfian sense, but by changing probabilistic inference, category boundaries, and similarity structure. Second, the **conceptual spaces** framework in cognitive science: concepts are regions in a structured space spanned by interpretable dimensions such as hue, brightness, pitch, or risk, where similarity is literally distance and prototypes sit near the centers of those regions. Third, the **embedding machinery** of contemporary NLP: learned high-dimensional spaces in which semantic similarity behaves like distance and linear structure supports analogies and control. GWH treats LLM embeddings as language-specific and context-specific geometries over a shared manifold of meanings, constrained and evaluated against human cognitive data.</p><p>From this perspective, linguistic relativity becomes *explicitly* geometric. The categories your language gives you, for example color terms, spatial frames, and number systems, do not merely label a conceptual space. They can actually warp it. Perceptual color space is approximately continuous and low-dimensional, yet languages impose different boundaries and distort perceived distances. Speakers with a single term for blue and green cluster colors differently from speakers who sharply distinguish “blue” from “green.” In GWH, that is not a loose metaphor. The underlying manifold is the same, but languages induce different metrics and partitions, so distances, neighborhoods, and geodesics through conceptual space become language dependent.</p><p>Riemannian geometry enters when you take factors such as context and emotion seriously. Semantic distance in human cognition is not uniform. Under financial anxiety, “risk” is pulled toward “loss” and away from “opportunity.” For a trader, “risk” is near “volatility” and “leverage.” For a clinician, it is near “symptoms” and “interventions.” Under moral threat, “contamination,” “impurity,” and “danger” can collapse into a tight cluster. In neutral reflection, they spread apart again. That looks like curvature: regions where the space is stretched, compressed, or bent by affect, expertise, and social framing. The manifold of possible meanings is shared, but emotional states, roles, and cultures induce different local geometries on it. In other words, they induce different metrics and curvatures. In LLM terms, context is not just “more tokens in the window.” It is a temporary modification of the metric on the semantic manifold that changes which directions are cheap to move in and which distinctions are sharpened or blurred.</p><p>For LLM research, this reframes several things. Similarity becomes explicitly metric conditioned, varying with language and context rather than being a single global cosine. Human cognitive data, for example similarity judgments, category boundaries, reaction times, and framing effects, become constraints on the learned metric rather than just downstream evaluation signals. Multilingual models stop being “one big space for all tokens” and become language-conditioned geometries over a shared manifold, with cross-lingual alignment defined as learning maps between differently curved spaces. Prompting, persona conditioning, and safety steering can be described as geometric operations: moving the current point, modifying the local metric, or constraining the allowed paths through the space.</p><p>The payoffs are concrete. For cross-lingual and cross-cultural alignment, you can explicitly map how concepts such as “harm,” “fairness,” or “consent” sit in differently curved spaces across languages, then design geometric transformations between them. For interpretability and safety, you can track how the neighborhoods around value-laden concepts shift under different prompts, user personas, or fine-tunes, and you can detect when adversarial prompting flattens important moral distinctions. For model design, you can encourage low-dimensional, interpretable submanifolds in domains such as color, space, or risk, and you can regularize the metric toward known psychophysical geometries instead of hoping they emerge from scale.</p><p>The GWH is a research program rather than a finished theory. Many of the implications will be seen as controversial, although the program is grounded in established and widely accepted work in cognitive linguistics, conceptual spaces, differential geometry, and recent AI research. The goal is to build tools that map “semantic ecologies,” probe and reshape LLM embedding spaces using human data as geometric supervision, and design interfaces where people can inspect how a model currently “sees” a domain and nudge that geometry toward healthier forms. The core thesis is straightforward: if language shapes thought, it does so by shaping the space in which thinking happens, and LLMs are our best current approximations to those spaces. GWH is about learning to read, compare, and carefully redesign that geometry so that human minds and machine models inhabit compatible, inspectable conceptual worlds. The internal models that represent both the external world and the observing and acting agent are plausible core components of consciousness in humans and other biological organisms, and the same style of internal modeling offers a route to analogous structures in artificial intelligences and non-organic life as well.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a2a0aac4180f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian/the-accidental-geometer-the-geometric-whorfian-hypothesis-for-llms-a2a0aac4180f">The Accidental Geometer — The Geometric Whorfian Hypothesis for LLMs</a> was originally published in <a href="https://medium.com/the-accidental-geometer-the-geometric-whorfian">emergenesis</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>