# L"ushi-1840: Building and Probing a Knowledge Cutoff for a Classical Chinese Language Model

arXiv: https://arxiv.org/abs/2610.16201 · HTML: https://mlsanity.com/paper/2610.16201/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 14 of its 34 references that are in the map.
- First posted on arXiv 12 Oct 2026.

## Builds on (most cited first, 14 of 14)

- [Time-Aware Language Models as Temporal Knowledge Bases](https://mlsanity.com/paper/2106.15110/) (2021)
- [Mind the Gap: Assessing Temporal Generalization in Neural Language Models](https://mlsanity.com/paper/2102.01951/) (2021)
- [UER: An Open-Source Toolkit for Pre-training Models](https://mlsanity.com/paper/1909.05658/) (2019)
- [Goldfish: Monolingual Language Models for 350 Languages](https://mlsanity.com/paper/2408.10441/) (2024)
- [Can Language Models Represent the Past without Anachronism?](https://mlsanity.com/paper/2505.00030/) (2025)
- [WYWEB: A NLP Evaluation Benchmark For Classical Chinese](https://mlsanity.com/paper/2305.14150/) (2023)
- [Time-Aware Ancient Chinese Text Translation and Inference](https://mlsanity.com/paper/2107.03179/) (2021)
- [SikuGPT: A Generative Pre-trained Model for Intelligent Information Processing of Ancient Texts from the Perspective of Digital Humanities](https://mlsanity.com/paper/2304.07778/) (2023)
- [DatedGPT: Preventing Lookahead Bias in Large Language Models with Time-Aware Pretraining](https://mlsanity.com/paper/2603.11838/) (2026)
- [Pretraining Language Models for Diachronic Linguistic Change Discovery](https://mlsanity.com/paper/2504.05523/) (2025)
- [WenyanGPT: A Large Language Model for Classical Chinese Tasks](https://mlsanity.com/paper/2504.20609/) (2025)
- [Internal Knowledge Without External Expression: Probing the Generalization Boundary of a Classical Chinese Language Model](https://mlsanity.com/paper/2604.14180/) (2026)
- [TimeCapsule: Generative Hallucination as a Method for Historical Sensemaking](https://mlsanity.com/paper/2607.24750/) (2026)
- [Do Language Models Consistently Encode the Current Year?](https://mlsanity.com/paper/2608.15507/) (2026)

## Abstract (arXiv)

Language models trained only on text from before a given year are useful for studying hindsight and historical language, but the open collections of pre-modern Chinese carry no publication dates and unreliable dynasty labels. We present L"ushi-1840, a 983M-parameter model trained from scratch on 12,262 books (1.28B characters) selected as composed before 1840. A first corpus, dated by a common heuristic, in effect ended at 1644 while containing twentieth-century material. We describe the re-dating and cleaning pipeline that replaced it and an audit with markers the pipeline did not use, which still finds later material. On the model side we measure surprisal, relative to a modern reference, on annals compiled after the cutoff. For events after 1840 against events up to 1820 the ratio is 1.38 (95% interval 1.33-1.44), against about 1.00 for modern general-purpose models. In a series of 294M models, adding the later court records lowers the ratio from 1.33 to 1.16, and the two models diverge from the 1820s, where the added records begin. The probe tracks the coverage of sources and does not establish a single cutoff year; none of our probes exposes the audited contamination. Model, book list and scripts will be released.