Copertina

Chain-of-thought obfuscation learned from output supervision can generalise to unseen tasks

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Hadida, Nathaniel Mitrani, Bhanji, Sassan, Tice, Cameron, Radmard, Puria
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

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

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

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

Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment
di: Tice, Cameron, et al.
Pubblicazione: (2026)

Behavioural Analysis of Alignment Faking
di: Hadida, Nathaniel Mitrani, et al.
Pubblicazione: (2026)

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

M3Kang: Evaluating Multilingual Multimodal Mathematical Reasoning in Vision-Language Models
di: Torres-Camps, Aleix, et al.
Pubblicazione: (2026)

Backdoor defense, learnability and obfuscation
di: Christiano, Paul, et al.
Pubblicazione: (2024)

Chaining thoughts and LLMs to learn DNA structural biophysics
di: Ross, Tyler D., et al.
Pubblicazione: (2024)

Detecting new obfuscated malware variants: A lightweight and interpretable machine learning approach
di: Madamidola, Oladipo A., et al.
Pubblicazione: (2024)

Weakly supervised deep learning model with size constraint for prostate cancer detection in multiparametric MRI and generalization to unseen domains
di: Trombetta, Robin, et al.
Pubblicazione: (2024)

Out-of-distribution generalisation is hard: evidence from ARC-like tasks
di: Dimitriadis, George, et al.
Pubblicazione: (2025)

LLMs can construct powerful representations and streamline sample-efficient supervised learning
di: Demirel, Ilker, et al.
Pubblicazione: (2026)

An exploration of features to improve the generalisability of fake news detection models
di: Hoy, Nathaniel, et al.
Pubblicazione: (2025)

Bayesian uncertainty-weighted loss for improved generalisability on polyp segmentation task
di: Stone, Rebecca S., et al.
Pubblicazione: (2023)

MuCRASP: Multimodal Chain-of-thought Reasoning aware Structured Pruning
di: Dutta, Aritra, et al.
Pubblicazione: (2026)

Chain or tree? Re-evaluating complex reasoning from the perspective of a matrix of thought
di: Tang, Fengxiao, et al.
Pubblicazione: (2025)

Multi-task learning on partially labeled datasets via invariant/equivariant semi-supervised learning
di: Rabadán, Miquel Martí i, et al.
Pubblicazione: (2026)

Improve Vision Language Model Chain-of-thought Reasoning
di: Zhang, Ruohong, et al.
Pubblicazione: (2024)

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

To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning
di: Sprague, Zayne, et al.
Pubblicazione: (2024)

SketchVLM: Vision language models can annotate images to explain thoughts and guide users
di: Collins, Brandon, et al.
Pubblicazione: (2026)

PKRD-CoT: A Unified Chain-of-thought Prompting for Multi-Modal Large Language Models in Autonomous Driving
di: Luo, Xuewen, et al.
Pubblicazione: (2024)

Comparing supervised learning dynamics: Deep neural networks match human data efficiency but show a generalisation lag
di: Huber, Lukas S., et al.
Pubblicazione: (2024)

Chain-of-Description: What I can understand, I can put into words
di: Guo, Jiaxin, et al.
Pubblicazione: (2025)

Curriculum reinforcement learning with measurable task representation learning
di: Wen, Yongyan, et al.
Pubblicazione: (2026)

Humans can learn to detect AI-generated texts, or at least learn when they can't
di: Milička, Jiří, et al.
Pubblicazione: (2025)

PACAD-Based Structural Reasoning for AGI: From Probabilistic Pattern-Matching to Verifiable Canonical Architectures – paper 1, version 2
di: Brown, Cameron
Pubblicazione: (2026)

Code-enabled language models can outperform reasoning models on diverse tasks
di: Zhang, Cedegao E., et al.
Pubblicazione: (2025)

chat log: Goolge Search AI - Anthropic Claude benchmarks and model release statistics in comparison to PACAD-based training estimates
di: Brown, Cameron
Pubblicazione: (2026)

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

PACAD-Enhanced Opus: Specialized Benchmark Performance Estimates – 2-3 Week Cycle (14-21 Days, Median ~17 Days)
di: Brown, Cameron
Pubblicazione: (2026)

BiRoDiff: Diffusion policies for bipedal robot locomotion on unseen terrains
di: Mothish, GVS, et al.
Pubblicazione: (2024)

AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning
di: Wang, Tevin, et al.
Pubblicazione: (2025)

PACAD-Based Structural Reasoning for AGI: From Probabilistic Pattern-Matching to Verifiable Canonical Architectures – paper 1, version 1
di: Brown, Cameron
Pubblicazione: (2026)

AEC — Axiomatic Engine Cycle v0.1 — Initial Specification Draft
di: Brown, Cameron
Pubblicazione: (2026)

Self-supervised learning on gene expression data
di: Dradjat, Kevin, et al.
Pubblicazione: (2025)

Towards efficient representation identification in supervised learning
di: Ahuja, Kartik, et al.
Pubblicazione: (2022)

Generalizable Chain-of-Thought Prompting in Mixed-task Scenarios with Large Language Models
di: Zou, Anni, et al.
Pubblicazione: (2023)

Generalization capabilities of MeshGraphNets to unseen geometries for fluid dynamics
di: Schmöcker, Robin, et al.
Pubblicazione: (2024)