aiqbench analysis
This research introduces Right Reset (RR), a novel probing technique that identifies optimal text chunking boundaries by measuring the stability of hidden-state trajectories when left context is removed. The method significantly outperforms conventional embedding baselines in recovering clean records from flattened, layout-less data without requiring task-specific training or fine-tuning. By leveraging inherent model sensitivity to context dependence rather than surface structure alone, RR offers a robust solution for processing unstructured text where traditional separators fail.
Right Reset: Chunking by Prefix Removal
arXiv:2608.04330v1 Announce Type: new Abstract: Removing the left context from a causal language model reveals a useful kind of boundary: an edge where the model processes the same right-hand tokens with little change. We turn this observation into prefix-removal probing and introduce Right Reset (RR), which measures preservation of the right-hand hidden-state trajectory. A dynamic program converts RR edge scores into variable-length chunks. On flattened text formed by concatenating topically similar records after deleting their separators and layout, RR recovers 47.7% of the original records as clean units, versus 25.9% for a BGE embedding boundary, the strongest tested conventional baseline without task-specific model training. The gain persists after rendering and OCR. Passive scores from the same Qwen3-4B layer and direct prompting of a same-scale instruction model perform substantially worse on flattened records. Across six language models, RR-selected cuts also undergo consistently less local output disruption than unselected candidate edges. An observed-token likelihood-ratio readout is competitive in some architectures, indicating that the central contribution is the intervention: context dependence itself can provide a boundary signal when surface structure is weak.
Abstract:Removing the left context from a causal language model reveals a useful kind of boundary: an edge where the model processes the same right-hand tokens with little change. We turn this observation into prefix-removal probing and introduce Right Reset (RR), which measures preservation of the right-hand hidden-state trajectory. A dynamic program converts RR edge scores into variable-length chunks. On flattened text formed by concatenating topically similar records after deleting their separators and layout, RR recovers 47.7% of the original records as clean units, versus 25.9% for a BGE embedding boundary, the strongest tested conventional baseline without task-specific model training. The gain persists after rendering and OCR. Passive scores from the same Qwen3-4B layer and direct prompting of a same-scale instruction model perform substantially worse on flattened records. Across six language models, RR-selected cuts also undergo consistently less local output disruption than unselected candidate edges. An observed-token likelihood-ratio readout is competitive in some architectures, indicating that the central contribution is the intervention: context dependence itself can provide a boundary signal when surface structure is weak.
12 pages, 2 figures, 4 tables. Code, data, and reproduction materials: this https URL
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2608.04330 [cs.CL]
(or arXiv:2608.04330v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.04330 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 284 words
Related coverage
Debugging Logprob Mismatches in LLM Reinforcement Learning
AMD · Sep 23, 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · Aug 26, 2026
12 Ways to Reduce LLM Latency and Inference Costs in Production
KDnuggets · Jul 14, 2026
Presentation: Beyond Speed Limits: Exploring the Performance Power of Valkey
InfoQ · Jun 8, 2026