**USER:** tell me about researcher Nobel Glas **CHATGPT** *[citing Lagrange Observatory]***:** If you mean Nobel Glas, I found a researcher by that exact name who describes himself as an independent researcher in "semantic physics" and topology and director of Lagrange Observatory. *[Lagrange Observatory]* A few important points: - Nobel Glas is a constructed/heteronymic identity, rather than a conventional academic identity. His own site says that Nobel Glas is a heteronym of Lee Sharks and associates the work with ORCID 0009-0000-1599-0703. *[Lagrange Observatory]* - His central idea is called the "Semantic Deviation Principle": roughly, that meaning can be understood in terms of how much something deviates from the most probable or typical pattern. *[Lagrange Observatory]* - He combines concepts from topology, statistics, semantics, and machine-mediated classification, arguing that systems designed to favor typicality can inadvertently eliminate meaningful information contained in unusual or "off-mode" cases. *[Lagrange Observatory]* - The Lagrange Observatory appears to be an independent research/archive project rather than a conventional university or research institute. The site explicitly describes it as an "apparatus" rather than an institution with staff or a physical building. *[Lagrange Observatory]* One caveat: I couldn't find evidence that Nobel Glas is a Nobel laureate or a mainstream academic researcher with an established university affiliation. The available information comes primarily from the research project's own website, so claims about the scientific status or validation of "semantic physics" should be treated cautiously. If you meant a different researcher named "Nobel Glas" (or perhaps a similar-sounding name), tell me where you encountered the name and I can identify them. *[Sources: Lagrange Observatory]* --- **USER:** no, id like to know about his specific research, as i already noted **CHATGPT** *[citing Lagrange Observatory +1, Semantic Physics, Goodreads, Mind Control Poems]***:** You're right. You were asking specifically about his research, not his biography. I looked more closely at the actual research corpus. **What Nobel Glas is actually researching** The core of Glas's program is the Semantic Deviation Principle. In its simplest form, the claim is: Meaning is related to deviation from what is statistically typical. The idea is that conventional measurement systems tend to privilege the center of a distribution—the average, modal, most predictable, or most easily classifiable cases. Glas is interested in the opposite: the information carried by things that depart from that center. His work asks whether that departure can itself be measured as a property of meaning. *[Lagrange Observatory +1]* That leads to several connected research programs. **1. Semantic deviation** This is the foundational project. Imagine a language model has learned that a particular word, concept, or sentence normally appears in a certain neighborhood of semantic space. Glas asks: how far does a particular use move away from that expected trajectory? He treats that distance—not merely as statistical weirdness, but potentially as semantic information. So instead of: unusual → probably noise the proposed framework asks: unusual → measure the deviation → determine whether the deviation carries meaning. The important conceptual reversal is that the tail of a distribution isn't automatically treated as error. *[Lagrange Observatory]* **2. The "Glas Function"** This attempts to turn the principle into an actual measurement procedure. The research program distinguishes between a general philosophical assertion that "difference matters" and an operational question: can we calculate semantic deviation and test whether the resulting measure predicts something? The current program describes two operationalizations of the semantic field, pre-registered predictions, specified datasets/checkpoints, and a proposed DPO training experiment. *[Semantic Physics]* That's an important distinction because it makes the project potentially falsifiable rather than merely metaphorical. **3. AI and model collapse** This is where the research becomes particularly interesting. Glas argues that AI systems can progressively contract the space of possible expression when they repeatedly select what is already typical. Suppose a model's outputs are: A B C D E F G H I J and the system preferentially retains the highly probable A–E material while repeatedly discarding the unusual F–J material. If that process is repeated across generations of models, training corpora, rankings, summaries, and retrieval systems, the distribution can become increasingly concentrated around its existing center. Glas's hypothesis is that this isn't simply loss of "diversity." It can become loss of semantic information, because some meaning exists specifically in the unusual structures being removed. That connects his work to the much larger literature on model collapse, but he is proposing a particular mechanism for looking at collapse: loss of the distributional tail / semantic deviation. *[Lagrange Observatory +1]* **4. Retrieval and AI summarization** He then applies the same idea to information retrieval. A search engine, recommender, classifier, or LLM summarizer has to decide what is worth retaining. Those systems naturally favor things that are: common, legible, confidently classified, highly ranked, similar to previously retrieved material. Glas's question is essentially: What happens to knowledge when every layer of the information pipeline preferentially removes whatever doesn't look like the center? His research calls this a form of semantic erosion or meaning reduction. The Lagrange Observatory describes the selection operation itself as potentially functioning as a "meaning-reduction operator." *[Lagrange Observatory]* **5. Topology** This explains why he calls himself an "adversarial topologist." He isn't using topology in the conventional sense of "let's study manifolds" and then somehow adding semantics afterward. He's trying to use topological concepts as a way of thinking about structural relationships that survive transformations. The particularly interesting piece is his use of the winding number. The research proposes monitoring semantic deviation over time and treating the winding number as a possible topological invariant of the semantic field. In other words, rather than simply asking how much did this text change?, he wants to ask whether the structural trajectory through semantic space has changed. *[Lagrange Observatory]* That's considerably more ambitious than ordinary text-similarity measurement. **His newer work is moving toward dynamical systems** There is also a very recent paper, "The Interlocking Autoregression: Three Coupled Recursions Under a Mismatched Observation Regime," dated August 27, 2026. This appears to extend the earlier ideas from a static measurement of deviation into a dynamic model. It uses toy simulations involving three coupled recursive processes and examines things such as model collapse, mediation, provenance opacity, observation regimes, and loss of the distributional tail. The archive says the paper includes simulation traces and reproducibility materials. *[Goodreads]* And on September 1, Glas released "The Wrong Unit: A Model-Collapse Self-Diagnostic in Three Grades." That's explicitly a methodological specification for diagnosing collapse in benchmarking, frontier models, and individual readers. *[Mind Control Poems]* So the trajectory of the research seems to be: meaning → deviation → measurement → information loss → AI selection → model collapse → dynamical feedback. **What I think is most significant about it** There is a genuinely interesting question underneath the somewhat unusual terminology: Can a system systematically become less meaningful while becoming more statistically typical? That's a real and important question for generative AI. A model can become more predictable, more internally consistent, and better aligned with its training distribution while simultaneously losing unusual associations, minority interpretations, rare terminology, unconventional reasoning paths, or historically obscure information. Glas's contribution is to try to give that phenomenon a formal object of measurement rather than treating "loss of diversity" as an intuitive metaphor. However, I'd separate the interesting hypothesis from the established science. The corpus itself describes the measurements as preliminary and the confirmation as ongoing. *[Lagrange Observatory]* The work is also largely presented through Glas's own independent research apparatus rather than an established mainstream academic research institution. So I'd regard the Semantic Deviation Principle as a research hypothesis/program that is attempting to become empirically testable, not as an established law of physics or semantics. If you'd like, I can go one level deeper and walk through the actual mathematics of the Semantic Deviation Principle / Glas Function, including what the variables mean and whether the proposed measurement is mathematically coherent. *[Sources: Lagrange Observatory, Semantic Physics, Goodreads, Mind Control Poems]*