analiza aiqbench
This research introduces a novel method for optimizing document suffixes to compel language models to adhere to provided context, even when it contradicts their parametric knowledge. By using discrete token search without direct model access or explicit instructions, the technique effectively overrides model resistance in counterfactual scenarios and generalizes across unseen passages. It demonstrates that carefully crafted document extensions can achieve performance comparable to activation steering and strong manual instructions, offering a new approach for improving context following in RAG systems.
Shaping the Document, Not the Model: Optimized Suffixes for Context Following
arXiv:2610.26236v1 Announce Type: new Abstract: When a user supplies a document and wants the answer drawn from it, a language model is expected to follow it even when it contradicts its parametric knowledge. Prior work has focused on making the model rely on context by steering activations, contrasting logits during decoding, or instructing it in plain language. We ask whether the document alone, with no instruction written into it and no access to the model at inference time, is enough. We optimize a document suffix by discrete token search, starting from the answer to a single question and extending the objective to a set of passages, to multi-hop questions that follow from the planted fact, and to later turns of a conversation. On counterfactual passages that contradict parametric knowledge of the model, the suffix flips nearly all items the model resists across three instruction-tuned model families, in most settings outperforms the decoding baseline and the strongest instruction we could write, and matches activation steering. Controls show that the suffix largely acts through the passage rather than injecting the answer, and a single suffix per model transfers to unseen passages. Each extended objective carries the effect to its target, so the objective, not only the document, sets how far the effect reaches.
Abstract:When a user supplies a document and wants the answer drawn from it, a language model is expected to follow it even when it contradicts its parametric knowledge. Prior work has focused on making the model rely on context by steering activations, contrasting logits during decoding, or instructing it in plain language. We ask whether the document alone, with no instruction written into it and no access to the model at inference time, is enough. We optimize a document suffix by discrete token search, starting from the answer to a single question and extending the objective to a set of passages, to multi-hop questions that follow from the planted fact, and to later turns of a conversation. On counterfactual passages that contradict parametric knowledge of the model, the suffix flips nearly all items the model resists across three instruction-tuned model families, in most settings outperforms the decoding baseline and the strongest instruction we could write, and matches activation steering. Controls show that the suffix largely acts through the passage rather than injecting the answer, and a single suffix per model transfers to unseen passages. Each extended objective carries the effect to its target, so the objective, not only the document, sets how far the effect reaches.
19 pages, 5 figures, 18 tables. Project page: this https URL
Computation and Language (cs.CL)
arXiv:2610.26236 [cs.CL]
(or arXiv:2610.26236v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.26236 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Text integral sub licența deschisă a publicației · 312 cuvinte
Subiecte conexe
Graph4BiLO: Graph Neural Network Approximation for Bilevel Mixed-Integer Linear Optimization
arXiv · 1 sept. 2026
Optimize blueprint extraction accuracy in Amazon Bedrock Data Automation
AWS · 11 iun. 2026
Sample-Efficient Hypergradient Estimation for Decentralized Bi-Level Reinforcement Learning
arXiv · 12 iun. 2026
Skill$^{\infty}$: Recursive Self-Improvement for Agent Skill Optimization
arXiv · 27 oct. 2026