Paper Overview
Field: Machine Learning Authors: Reuben Vandeventer, David Imrem, David J. Wild Published: 2026-09-15 arXiv: 2609.12105
Summary
The prevailing assumption in applied machine learning is that progress on consequential quantitative decisions such as pricing risk, allocating capital, triaging patients, or containing a network intrusion will follow from progress in large language models (LLMs). A language model is trained on a representation of the world that was produced by human description; description is a lossy encoding of the quantitative record, and the loss is irreversible: no downstream model, at any scale, can recover from a description what the description did not encode. We formalize this as a property of the representation on which a model is trained rather than of the model capacity, and we identify three further properties that consequential settings demand of a model and that a language substrate cannot provide by construction: reproducibility, lineage from each output back to the source records that produced it, and calibrated uncertainty. We argue that these properties define a distinct class of model, which we call the Large Quantitative Model (LQM).
Key Takeaways
- Description of quantitative records is a lossy, irreversible encoding — no downstream model of any scale can recover information that was never encoded.
- The limitation is framed as a property of the training representation, not of model capacity.
- Three properties demanded by high-stakes settings but unavailable from a language substrate: 1. Reproducibility 2. Lineage — traceability of each output to its source records 3. Calibrated uncertainty
- The authors propose a new model category: the Large Quantitative Model (LQM).