Cover Image

Interpretability without actionability: mechanistic methods cannot correct language model errors despite near-perfect internal representations

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Basu, Sanjay, Patel, Sadiq Y., Sheth, Parth, Muralidharan, Bhairavi, Elamaran, Namrata, Kinra, Aakriti, Morgan, John, Batniji, Rajaie
Format:	Preprint
Published:	2026
Subjects:	Artificial Intelligence I.2.7; J.3
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Feasibility-Guided Fair Adaptive Offline Reinforcement Learning for Medicaid Care Management
by: Basu, Sanjay, et al.
Published: (2025)

Hybrid Adaptive Conformal Offline Reinforcement Learning for Fair Population Health Management
by: Basu, Sanjay, et al.
Published: (2025)

Test-Time Learning and Inference-Time Deliberation for Efficiency-First Offline Reinforcement Learning in Care Coordination and Population Health Management
by: Basu, Sanjay, et al.
Published: (2025)

Comparing Alternative Approaches to Care Management Prioritization: A Prospective Comparative Cohort Study of Acute Care Utilization and Equity Among Medicaid Beneficiaries
by: Parth Sheth, et al.
Published: (2026)

LLMs cannot find reasoning errors, but can correct them given the error location
by: Tyen, Gladys, et al.
Published: (2023)

On Continuous Data Assimilation for a class of 2D and 3D stochastic non-Newtonian fluids of differential type
by: Kinra, Kush
Published: (2026)

A note on continuous data assimilation for stochastic convective Brinkman-Forchheimer equations in 2D and 3D
by: Kinra, Kush
Published: (2026)

Writing Self, Writing Empire
by: Kinra, Rajeev
Published: (2020)

Writing Self, Writing Empire: Chandar Bhan Brahman and the Cultural World of the Indo-Persian State Secretary
by: Kinra, Rajeev
Published: (2016)

Rate of convergence of random attractors towards deterministic singleton attractor for a class of non-Newtonian fluids of differential type
by: Kinra, Kush
Published: (2026)

Large time behaviour for a class of 2D and 3D stochastic non-Newtonian fluids of differential types: Attractors and invariant measures
by: Kinra, Kush
Published: (2026)

Machine Learning Invariants of Tensors
by: Elamaran, Athithan, et al.
Published: (2025)

Revisiting representations of quark mixing matrix
by: Kaur, Gurjit, et al.
Published: (2023)

Locally analytic representations in the étale coverings of the Lubin-Tate moduli space
by: Sheth, Mihir
Published: (2017)

The Value Sensitivity Gap: How Clinical Large Language Models Respond to Patient Preference Statements in Shared Decision-Making
by: Basu, Sanjay
Published: (2026)

Product representation of perfect cubes
by: Fleiner, Zsigmond György, et al.
Published: (2024)

Product representations of perfect powers
by: Pach, Péter Pál, et al.
Published: (2026)

Is plantar thermography a valid digital biomarker for characterising diabetic foot ulceration risk?
by: Jagadeesh, Akshay, et al.
Published: (2024)

First-order ANIL provably learns representations despite overparametrization
by: Yüksel, Oğuz Kaan, et al.
Published: (2023)

Global-in-time optimal control of stochastic third-grade fluids with additive noise
by: Kinra, Kush, et al.
Published: (2025)

Random dynamics of solutions for three-dimensional stochastic globally modified Navier-Stokes equations on unbounded Poincaré domains
by: Kinra, Kush, et al.
Published: (2024)

Stochastic convective Brinkman-Forchheimer equations on general Unbounded Domains
by: Kinra, Kush, et al.
Published: (2020)

Random dynamics and invariant measures for a class of non-Newtonian fluids of differential type on 2D and 3D Poincaré domains
by: Kinra, Kush, et al.
Published: (2025)

Backward uniqueness of 2D and 3D convective Brinkman-Forchheimer equations and its applications
by: Kinra, Kush, et al.
Published: (2023)

Non-uniqueness of Hölder continuous solutions for stochastic Euler and Hypodissipative Navier-Stokes equations
by: Kinra, Kush, et al.
Published: (2024)

Optimal control problem associated with three-dimensional critical convective Brinkman-Forchheimer equations
by: Kinra, Kush, et al.
Published: (2026)

Combinatorial generation via permutation languages. VI. Binary trees
by: Gregor, Petr, et al.
Published: (2023)

Division Strategies Determine Growth and Viability of Growing-Dividing Autocatalytic Systems
by: Pandey, Parth Pratim, et al.
Published: (2025)

On the integrality of locally algebraic representations of $\mathrm{GL}_{2}(D)$
by: Nadimpalli, Santosh, et al.
Published: (2023)

Measurement-free fault-tolerant quantum error correction in near-term devices
by: Heußen, Sascha, et al.
Published: (2023)

Brain alignment of reasoning and action representations from vision-language and action models during naturalistic gameplay
by: Oota, Subba Reddy, et al.
Published: (2026)

“English is important, but perfect English is not”: The disjuncture between the IELTS and language experiences from the perspectives of international students in Australia
by: Xiaofan Liu, et al.
Published: (2024)

A global AI community requires language-diverse publishing
by: Lepp, Haley, et al.
Published: (2024)

LLMs cannot spot math errors, even when allowed to peek into the solution
by: Srivatsa, KV Aditya, et al.
Published: (2025)

Qudit-based quantum error-correcting codes from irreducible representations of SU(d)
by: Uy, Robert Frederik, et al.
Published: (2024)

Proof of a perfect platonic representation hypothesis
by: Ziyin, Liu, et al.
Published: (2025)

A mechanistic model for smallpox transmission via inhaled aerosols inside respiratory pathways
by: Basu, Saikat, et al.
Published: (2024)

Large Language Model probabilities cannot distinguish between possible and impossible language
by: Leivada, Evelina, et al.
Published: (2025)

Relativistic perfect fluids near Kasner singularities
by: Beyer, Florian, et al.
Published: (2020)

Prompting open-source and commercial language models for grammatical error correction of English learner text
by: Davis, Christopher, et al.
Published: (2024)