Cover Image

Position: Capability Control Should be a Separate Goal From Alignment

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Siddiqui, Shoaib Ahmed, Triantafillou, Eleni, Krueger, David, Weller, Adrian
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2025)

On Evaluating LLMs' Capabilities as Functional Approximators: A Bayesian Perspective
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2024)

Blockwise Self-Supervised Learning at Scale
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2023)

A deeper look at depth pruning of LLMs
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2024)

Exploring the design space of deep-learning-based weather forecasting systems
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2024)

The Topological Trouble With Transformers
by: Mozer, Michael C., et al.
Published: (2026)

Permissive Information-Flow Analysis for Large Language Models
by: Siddiqui, Shoaib Ahmed, et al.
Published: (2024)

Position: AI Evaluations Should be Grounded on a Theory of Capability
by: Jo, Nathanael, et al.
Published: (2025)

Step-resolved data attribution for looped transformers
by: Kaissis, Georgios, et al.
Published: (2026)

Your Privacy Depends on Others: Collusion Vulnerabilities in Individual Differential Privacy
by: Kaiser, Johannes, et al.
Published: (2026)

Estimation of Concept Explanations Should be Uncertainty Aware
by: Piratla, Vihari, et al.
Published: (2023)

Protecting against simultaneous data poisoning attacks
by: Alex, Neel, et al.
Published: (2024)

Position: Weight Space Should Be a First-Class Generative AI Modality
by: Wang, Zhangyang, et al.
Published: (2026)

To Each (Textual Sequence) Its Own: Improving Memorized-Data Unlearning in Large Language Models
by: Barbulescu, George-Octavian, et al.
Published: (2024)

On the Limitations and Capabilities of Position Embeddings for Length Generalization
by: Chen, Yang, et al.
Published: (2025)

Rotary Position Encodings for Graphs
by: Reid, Isaac, et al.
Published: (2025)

Position: Beyond Euclidean -- Foundation Models Should Embrace Non-Euclidean Geometries
by: He, Neil, et al.
Published: (2025)

GRAIL: Goal Recognition Alignment through Imitation Learning
by: Elhadad, Osher, et al.
Published: (2026)

ALVIN: Active Learning Via INterpolation
by: Korakakis, Michalis, et al.
Published: (2024)

Mitigating Shortcut Learning with InterpoLated Learning
by: Korakakis, Michalis, et al.
Published: (2025)

Position: AI Security Policy Should Target Systems, Not Models
by: Riegler, Michael A., et al.
Published: (2026)

Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
by: Song, Xiangchen, et al.
Published: (2025)

Getting By Goal Misgeneralization With a Little Help From a Mentor
by: Trinh, Tu, et al.
Published: (2024)

1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities
by: Wang, Kevin, et al.
Published: (2025)

From Tokenizer Bias to Backbone Capability: A Controlled Study of LLMs for Time Series Forecasting
by: Zhang, Xinyu, et al.
Published: (2025)

You Don't Need All That Attention: Surgical Memorization Mitigation in Text-to-Image Diffusion Models
by: Zhao, Kairan, et al.
Published: (2026)

Controlling Grokking with Nonlinearity and Data Symmetry
by: Salah, Ahmed, et al.
Published: (2024)

Noisy-Pair Robust Representation Alignment for Positive-Unlabeled Learning
by: Zhao, Hengwei, et al.
Published: (2025)

The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment
by: Balasubramanian, Rishab, et al.
Published: (2026)

Why Deep Jacobian Spectra Separate: Depth-Induced Scaling and Singular-Vector Alignment
by: Haas, Nathanaël, et al.
Published: (2026)

Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings
by: Lagesse, Adrien, et al.
Published: (2025)

Why Goal-Conditioned Reinforcement Learning Works: Relation to Dual Control
by: Lawrence, Nathan P., et al.
Published: (2025)

Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies
by: Zhao, Guangyu, et al.
Published: (2024)

Position: Machine Learning Conferences Should Establish a "Refutations and Critiques" Track
by: Schaeffer, Rylan, et al.
Published: (2025)

Token Reduction Should Go Beyond Efficiency in Generative Models -- From Vision, Language to Multimodality
by: Kong, Zhenglun, et al.
Published: (2025)

How Should We Meta-Learn Reinforcement Learning Algorithms?
by: Goldie, Alexander David, et al.
Published: (2025)

Learning to Decide with AI Assistance under Human-Alignment
by: Benz, Nina Corvelo, et al.
Published: (2026)

Dense and Diverse Goal Coverage in Multi Goal Reinforcement Learning
by: Singh, Sagalpreet, et al.
Published: (2025)

Implicit meta-learning may lead language models to trust more reliable sources
by: Krasheninnikov, Dmitrii, et al.
Published: (2023)

Autonomous Learning From Success and Failure: Goal-Conditioned Supervised Learning with Negative Feedback
by: Zhang, Zeqiang, et al.
Published: (2025)