Cover Image

Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Tice, Cameron, Radmard, Puria, Ratnam, Samuel, Kim, Andy, Africa, David, O'Brien, Kyle
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Chain-of-thought obfuscation learned from output supervision can generalise to unseen tasks
by: Hadida, Nathaniel Mitrani, et al.
Published: (2026)

A transformer architecture alteration to incentivise externalised reasoning
by: Pavlova, Elizabeth, et al.
Published: (2026)

Large language models can learn and generalize steganographic chain-of-thought under process supervision
by: Skaf, Joey, et al.
Published: (2025)

Diagnosing Pathological Chain-of-Thought in Reasoning Models
by: Liu, Manqing, et al.
Published: (2026)

Few-shot Personalization of LLMs with Mis-aligned Responses
by: Kim, Jaehyung, et al.
Published: (2024)

Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors
by: Radmard, Puria, et al.
Published: (2025)

Learning Dynamics of Meta-Learning in Small Model Pretraining
by: Africa, David Demitri, et al.
Published: (2025)

Meta-Pretraining for Zero-Shot Cross-Lingual Named Entity Recognition in Low-Resource Philippine Languages
by: Africa, David Demitri, et al.
Published: (2025)

The Curious Case of Factual (Mis)Alignment between LLMs' Short- and Long-Form Answers
by: Islam, Saad Obaid ul, et al.
Published: (2025)

Alignment-Weighted DPO: A principled reasoning approach to improve safety alignment
by: Hu, Mengxuan, et al.
Published: (2026)

FAIRE: Assessing Racial and Gender Bias in AI-Driven Resume Evaluations
by: Wen, Athena, et al.
Published: (2025)

LURE: Live-Usage Replay Evaluations for Reducing Evaluation Awareness
by: Ivanov, Igor, et al.
Published: (2026)

The Incomplete Bridge: How AI Research (Mis)Engages with Psychology
by: Jiang, Han, et al.
Published: (2025)

Steering Awareness: Detecting Activation Steering from Within
by: Rivera, Joshua Fonseca, et al.
Published: (2025)

MatheMagic: Generating Dynamic Mathematics Benchmarks Robust to Memorization
by: O'Brien, Dayyán, et al.
Published: (2025)

AI Managed Emergency Documentation with a Pretrained Model
by: Menzies, David, et al.
Published: (2024)

Privileged Self-Access Matters for Introspection in AI
by: Song, Siyuan, et al.
Published: (2025)

Language models align with human judgments on key grammatical constructions
by: Hu, Jennifer, et al.
Published: (2024)

The Great AI Witch Hunt: Reviewers Perception and (Mis)Conception of Generative AI in Research Writing
by: Hadan, Hilda, et al.
Published: (2024)

A flexible Bayesian non-parametric mixture model reveals multiple dependencies of swap errors in visual working memory
by: Radmard, Puria, et al.
Published: (2025)

Is It JUST Semantics? A Case Study of Discourse Particle Understanding in LLMs
by: Sheffield, William, et al.
Published: (2025)

Consistency Training while Mitigating Obfuscation via Rate Matching
by: Imran, Sohaib, et al.
Published: (2026)

Mis-prompt: Benchmarking Large Language Models for Proactive Error Handling
by: Zeng, Jiayi, et al.
Published: (2025)

(Mis)Fitting: A Survey of Scaling Laws
by: Li, Margaret, et al.
Published: (2025)

Human-Alignment and Calibration of Inference-Time Uncertainty in Large Language Models
by: Moore, Kyle, et al.
Published: (2025)

Investigating ReLoRA: Effects on the Learning Dynamics of Small Language Models
by: Weiss, Yuval, et al.
Published: (2025)

Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space
by: Ruscio, Valeria, et al.
Published: (2026)

Goal Alignment in LLM-Based User Simulators for Conversational AI
by: Mehri, Shuhaib, et al.
Published: (2025)

Association-sensory spatiotemporal hierarchy and functional gradient-regularised recurrent neural network with implications for schizophrenia
by: Abulikemu, Subati, et al.
Published: (2025)

Emergent Introspection in AI is Content-Agnostic
by: Lederman, Harvey, et al.
Published: (2026)

Fluent Alignment with Disfluent Judges: Post-training for Lower-resource Languages
by: Samuel, David, et al.
Published: (2025)

Sarc7: Evaluating Sarcasm Detection and Generation with Seven Types and Emotion-Informed Techniques
by: Xiong, Lang, et al.
Published: (2025)

Pause-Tuning for Long-Context Comprehension: A Lightweight Approach to LLM Attention Recalibration
by: Begin, James, et al.
Published: (2025)

Improving LLM Abilities in Idiomatic Translation
by: Donthi, Sundesh, et al.
Published: (2024)

Anchored Alignment for Self-Explanations Enhancement
by: Villa-Arenas, Luis Felipe, et al.
Published: (2024)

Code Pretraining Improves Entity Tracking Abilities of Language Models
by: Kim, Najoung, et al.
Published: (2024)

Reflection Pretraining Enables Token-Level Self-Correction in Biological Sequence Models
by: Zhang, Xiang, et al.
Published: (2025)

MisSynth: Improving MISSCI Logical Fallacies Classification with Synthetic Data
by: Poliakov, Mykhailo, et al.
Published: (2025)

Improving Dialogue Discourse Parsing through Discourse-aware Utterance Clarification
by: Fan, Yaxin, et al.
Published: (2025)

Causal Language Control in Multilingual Transformers via Sparse Feature Steering
by: Chou, Cheng-Ting, et al.
Published: (2025)