# Context-Parametric Inversion: Why Instruction Finetuning Can Worsen Context Reliance

arXiv: https://arxiv.org/abs/2410.10796 · HTML: https://mlsanity.com/paper/2410.10796/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Builds on 32 of its 40 references that are in the map.
- First posted on arXiv 14 Oct 2024.

## Builds on (most cited first, 30 of 32)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [REALM: Retrieval-Augmented Language Model Pre-Training](https://mlsanity.com/paper/2002.08909/) (2020)
- [Enhancing Chat Language Models by Scaling High-quality Instructional Conversations](https://mlsanity.com/paper/2305.14233/) (2023)
- [An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks](https://mlsanity.com/paper/1312.6211/) (2013)
- [An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning](https://mlsanity.com/paper/2308.08747/) (2023)
- [Dissecting Recall of Factual Associations in Auto-Regressive Language Models](https://mlsanity.com/paper/2304.14767/) (2023)
- [How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources](https://mlsanity.com/paper/2306.04751/) (2023)
- [Measuring Catastrophic Forgetting in Neural Networks](https://mlsanity.com/paper/1708.02072/) (2017)
- [Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts](https://mlsanity.com/paper/2305.13300/) (2023)
- [Trusting Your Evidence: Hallucinate Less with Context-aware Decoding](https://mlsanity.com/paper/2305.14739/) (2023)
- [RAFT: Adapting Language Model to Domain Specific RAG](https://mlsanity.com/paper/2403.10131/) (2024)
- [Inverse Scaling: When Bigger Isn't Better](https://mlsanity.com/paper/2306.09479/) (2023)
- [Large Language Models with Controllable Working Memory](https://mlsanity.com/paper/2211.05110/) (2022)
- [Evaluating Correctness and Faithfulness of Instruction-Following Models for Question Answering](https://mlsanity.com/paper/2307.16877/) (2023)
- [Rich Knowledge Sources Bring Complex Knowledge Conflicts: Recalibrating Models to Reflect Conflicting Evidence](https://mlsanity.com/paper/2210.13701/) (2022)
- [Understanding Catastrophic Forgetting in Language Models via Implicit Inference](https://mlsanity.com/paper/2309.10105/) (2023)
- [Tug-of-War Between Knowledge: Exploring and Resolving Knowledge Conflicts in Retrieval-Augmented Language Models](https://mlsanity.com/paper/2402.14409/) (2024)
- [Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Models](https://mlsanity.com/paper/2402.18154/) (2024)
- [Context-faithful Prompting for Large Language Models](https://mlsanity.com/paper/2303.11315/) (2023)
- [What does the Knowledge Neuron Thesis Have to do with Knowledge?](https://mlsanity.com/paper/2405.02421/) (2024)
- [Detecting and Mitigating Hallucinations in Multilingual Summarisation](https://mlsanity.com/paper/2305.13632/) (2023)
- [Two-stage LLM Fine-tuning with Less Specialization and More Generalization](https://mlsanity.com/paper/2211.00635/) (2022)
- [AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledge](https://mlsanity.com/paper/2409.07394/) (2024)
- [Understanding Finetuning for Factual Knowledge Extraction](https://mlsanity.com/paper/2406.14785/) (2024)
- [Discerning and Resolving Knowledge Conflicts through Adaptive Decoding with Contextual Information-Entropy Constraint](https://mlsanity.com/paper/2402.11893/) (2024)

## Built on by (most cited first, 10 of 10)

- [Mitigating Modal Imbalance in Multimodal Reasoning](https://mlsanity.com/paper/2510.02608/) (2025)
- [Can LLMs Reconcile Knowledge Conflicts in Counterfactual Reasoning](https://mlsanity.com/paper/2506.15732/) (2025)
- [Distilled Pretraining: A modern lens of Data, In-Context Learning and Test-Time Scaling](https://mlsanity.com/paper/2509.01649/) (2025)
- [Beyond Black-Box Interventions: Latent Probing for Faithful Retrieval-Augmented Generation](https://mlsanity.com/paper/2510.12460/) (2025)
- [Persona Conditioning of Brand Recommendations in Retrieval-Augmented Commercial Chat: A Prominence-Stratified Cross-Provider Audit](https://mlsanity.com/paper/2605.30207/) (2026)
- [Divergent Recommendations, Convergent Diagnoses: Cross-Provider Failure-Mode Convergence in AI Commercial Recommendation](https://mlsanity.com/paper/2606.26116/) (2026)
- [Do SpeechLMs Hear Their Own Opinions? Diagnosing and Mitigating Previous-Belief Contamination in Streaming Emotion Understanding](https://mlsanity.com/paper/2608.20769/) (2026)
- [ClimaEmpact: Domain-Aligned Small Language Models and Datasets for Extreme Weather Analytics](https://mlsanity.com/paper/2504.19066/) (2025)
- [Instruction Tuning with and without Context: Behavioral Shifts and Downstream Impact](https://mlsanity.com/paper/2506.15480/) (2025)
- [RAGStress: A controlled benchmark for evaluating retrieval-augmented generation under knowledge-base degradation](https://mlsanity.com/paper/2610.04691/) (2026)

## Abstract (arXiv)

A standard practice when using large language models is for users to supplement their instruction with an input context containing new information for the model to process. However, models struggle to reliably follow the input context, especially when it conflicts with their parametric knowledge from pretraining. In-principle, one would expect models to adapt to the user context better after instruction finetuning, particularly when handling knowledge conflicts. However, we observe a surprising failure mode: during instruction tuning, the context reliance under knowledge conflicts initially increases as expected, but then gradually decreases as instruction finetuning progresses. This happens while the performance on standard benchmarks keeps on increasing far after this drop. We call this phenomenon context-parametric inversion and observe it across multiple general purpose instruction tuning datasets such as TULU, Alpaca and Ultrachat, across different model families like Llama, Mistral, and Pythia. We perform various controlled studies and theoretical analysis to show that context-parametric inversion occurs due to examples in the instruction finetuning data where the input context provides information that aligns with model's parametric knowledge. Our analysis suggests some natural mitigation strategies with limited but insightful gains, and serves as a useful starting point in addressing this deficiency in instruction finetuning.