Image de couverture de livre

Large language models can learn and generalize steganographic chain-of-thought under process supervision

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Skaf, Joey, Ibanez-Lissen, Luis, McCarthy, Robert, Watts, Connor, Georgiv, Vasil, Whittingham, Hannes, Gonzalez-Manzano, Lorena, Lindner, David, Tice, Cameron, Young, Edward James, Radmard, Puria
Format:	Preprint
Publié:	2025
Sujets:	Artificial Intelligence Computation and Language Machine Learning
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Chain-of-thought obfuscation learned from output supervision can generalise to unseen tasks
par: Hadida, Nathaniel Mitrani, et autres
Publié: (2026)

Diagnosing Pathological Chain-of-Thought in Reasoning Models
par: Liu, Manqing, et autres
Publié: (2026)

A transformer architecture alteration to incentivise externalised reasoning
par: Pavlova, Elizabeth, et autres
Publié: (2026)

Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment
par: Tice, Cameron, et autres
Publié: (2026)

Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors
par: Radmard, Puria, et autres
Publié: (2025)

A flexible Bayesian non-parametric mixture model reveals multiple dependencies of swap errors in visual working memory
par: Radmard, Puria, et autres
Publié: (2025)

Association-sensory spatiotemporal hierarchy and functional gradient-regularised recurrent neural network with implications for schizophrenia
par: Abulikemu, Subati, et autres
Publié: (2025)

LPASS: Linear Probes as Stepping Stones for vulnerability detection using compressed LLMs
par: Ibanez-Lissen, Luis, et autres
Publié: (2025)

LUMIA: Linear probing for Unimodal and MultiModal Membership Inference Attacks leveraging internal LLM states
par: Ibanez-Lissen, Luis, et autres
Publié: (2024)

Esports can bring growth, but requires commitment
par: Claudine McCarthy
Publié: (2024)

Learn how positivity can improve work culture
par: Claudine McCarthy
Publié: (2025)

Learn how positivity can improve work culture
par: Claudine McCarthy
Publié: (2025)

Creative collaborations can support students’ mental health
par: Claudine McCarthy
Publié: (2024)

Creative collaborations can support students’ mental health
par: Claudine McCarthy
Publié: (2024)

Improved collaboration with ATs can boost student‐athlete experience
par: Claudine McCarthy
Publié: (2025)

Developing effective approaches to interpretations can help maintain compliance with NCAA rules
par: Claudine McCarthy
Publié: (2024)

Planning ahead can help protect your campus when natural disasters hit
par: Claudine McCarthy
Publié: (2025)

Planning ahead can help protect your campus when natural disasters hit
par: Claudine McCarthy
Publié: (2025)

Free expression teams can help protect students' rights while promoting campus safety
par: Claudine McCarthy
Publié: (2026)

Free expression teams can help protect students’ rights while promoting campus safety
par: Claudine McCarthy
Publié: (2026)

Shift in perspective can help DII, DIII schools tap into NIL as recruiting, retention tool
par: Claudine McCarthy
Publié: (2025)

Private key and password protection by steganographic image encryption
par: Choudhury, Debesh, et autres
Publié: (2025)

Las nuevas tecnologías conquistan las Universidades Andaluzas
par: Encarnación Sánchez Lissen
Publié: (2007)

La Vocación entre los Aspirantes a Maestro
par: Encarnación Sánchez Lissen
Publié: (2003)

Dos caras de la carrera docente: satisfacción y desmotivación
par: Encarnación Sánchez Lissen
Publié: (2009)

Salud personal y comunitaria: la influencia de nuestras actitudes y comportamientos ambientales
par: Encarnación Sánchez Lissen
Publié: (2005)

Quantum steganographic protocols using degenerate and entanglement-assisted quantum codes
par: Dutta, Sanjoy, et autres
Publié: (2025)

Towards Understanding Specification Gaming in Reasoning Models
par: Nishimura-Gasparian, Kei, et autres
Publié: (2026)

Dissociating language and thought in large language models
par: Mahowald, Kyle, et autres
Publié: (2023)

SketchVLM: Vision language models can annotate images to explain thoughts and guide users
par: Collins, Brandon, et autres
Publié: (2026)

The richness of language is the richness of thought
par: Hasanzoda, Rukhshona, et autres
Publié: (2026)

EGT Quantum Resonance: Predicting the $\mathbf{402 \text{ GeV}}$ WIMP and Future Discovery Conditions at the Large Hadron Collider
par: Tice, Brian
Publié: (2025)

Libertad de información
par: Jessica Whittingham
Publié: (2007)

Slow crossover from superdiffusion to diffusion in isotropic spin chains
par: McCarthy, Catherine, et autres
Publié: (2024)

Gravitational Baryogenesis in $f(R)$ Cosmologies
par: Whittingham, Ian B.
Publié: (2026)

Factores de riesgo relacionados con el Trabajo de Parto Prematuro en adolescentes embarazadas: revisión integradora de la literatura
par: Isabela Fleury Skaf Thomazini
Publié: (2016)

Using Neurogram Similarity Index Measure (NSIM) to Model Hearing Loss and Cochlear Neural Degeneration
par: Cheema, Ahsan J., et autres
Publié: (2025)

Utilizing Information Theoretic Approach to Study Cochlear Neural Degeneration
par: Cheema, Ahsan J., et autres
Publié: (2025)

How well can off-the-shelf LLMs elucidate molecular structures from mass spectra using chain-of-thought reasoning?
par: Wang, Yufeng, et autres
Publié: (2026)

Communist Multiculturalism
par: McCarthy, Susan
Publié: (2023)