# From Input Perception to Predictive Insight: Modeling Model Blind Spots Before They Become Errors

arXiv: https://arxiv.org/abs/2509.20065 · HTML: https://mlsanity.com/paper/2509.20065/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 58 references that are in the map.
- First posted on arXiv 24 Sep 2025.

## Builds on (most cited first, 22 of 22)

- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Discovering Latent Knowledge in Language Models Without Supervision](https://mlsanity.com/paper/2212.03827/) (2022)
- [The Internal State of an LLM Knows When It's Lying](https://mlsanity.com/paper/2304.13734/) (2023)
- [Transformer Interpretability Beyond Attention Visualization](https://mlsanity.com/paper/2012.09838/) (2020)
- [Eliciting Latent Predictions from Transformers with the Tuned Lens](https://mlsanity.com/paper/2303.08112/) (2023)
- [Regularizing Neural Networks by Penalizing Confident Output Distributions](https://mlsanity.com/paper/1701.06548/) (2017)
- [Analyzing the Structure of Attention in a Transformer Language Model](https://mlsanity.com/paper/1906.04284/) (2019)
- [Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Maps](https://mlsanity.com/paper/2407.07071/) (2024)
- [Testing the Predictions of Surprisal Theory in 11 Languages](https://mlsanity.com/paper/2307.03667/) (2023)
- [PUB: A Pragmatics Understanding Benchmark for Assessing LLMs' Pragmatics Capabilities](https://mlsanity.com/paper/2401.07078/) (2024)
- [Detecting Hallucinations in Large Language Model Generation: A Token Probability Approach](https://mlsanity.com/paper/2405.19648/) (2024)
- [Transformer-Based Language Model Surprisal Predicts Human Reading Times Best with About Two Billion Training Tokens](https://mlsanity.com/paper/2304.11389/) (2023)
- [On the Role of Context in Reading Time Prediction](https://mlsanity.com/paper/2409.08160/) (2024)
- [On the Effect of Anticipation on Reading Times](https://mlsanity.com/paper/2211.14301/) (2022)
- [Surprise! Uniform Information Density Isn't the Whole Story: Predicting Surprisal Contours in Long-form Discourse](https://mlsanity.com/paper/2410.16062/) (2024)
- [Sign of the Times: Evaluating the use of Large Language Models for Idiomaticity Detection](https://mlsanity.com/paper/2405.09279/) (2024)
- [Rolling the DICE on Idiomaticity: How LLMs Fail to Grasp Context](https://mlsanity.com/paper/2410.16069/) (2024)
- [Language Models Can Predict Their Own Behavior](https://mlsanity.com/paper/2502.13329/) (2025)
- [Enhancing Idiomatic Representation in Multiple Languages via an Adaptive Contrastive Triplet Loss](https://mlsanity.com/paper/2406.15175/) (2024)
- [Improve Decoding Factuality by Token-wise Cross Layer Entropy of Large Language Models](https://mlsanity.com/paper/2502.03199/) (2025)
- [ConMeC: A Dataset for Metonymy Resolution with Common Nouns](https://mlsanity.com/paper/2502.06087/) (2025)
- [Oddballness: universal anomaly detection with language models](https://mlsanity.com/paper/2409.03046/) (2024)

## Built on by (most cited first, 1 of 1)

- [Symbolic Augmentation Closes a Canonical-Equivalence Blind Spot in Neural Fact-Checkers](https://mlsanity.com/paper/2607.16212/) (2026)

## Abstract (arXiv)

Language models often struggle with idiomatic, figurative, or context-sensitive inputs, not because they produce flawed outputs, but because they misinterpret the input from the outset. We propose an input-only method for anticipating such failures using token-level likelihood features inspired by surprisal and the Uniform Information Density hypothesis. These features capture localized uncertainty in input comprehension and outperform standard baselines across five linguistically challenging datasets. We show that span-localized features improve error detection for larger models, while smaller models benefit from global patterns. Our method requires no access to outputs or hidden activations, offering a lightweight and generalizable approach to pre-generation error prediction.