Results 191 to 200 of about 21,797 (248)

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data

arXiv.org
We study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a"teacher"model with some trait T (such as liking owls or being misaligned) generates a ...
Alex Cloud, Minh Le, James Chua, Jan Betley, Anna Sztyber-Betley, Jacob Hilton, Samuel Marks, Owain Evans   +7 more
semanticscholar   +1 more source

Subliminal perception in cars

Proceedings of the 3rd International Conference on Automotive User Interfaces and Interactive Vehicular Applications, 2011
Following laws and provisions passed on the national and international level, the most relevant goal of future vehicular interfaces is to increase road safety. To alleviate the cognitive load associated with the interaction with the variety of emerging information and assistance systems in the car (and to increase driving performance as well ...
Andreas Riener, Myounghoon Jeon 0001, Manfred Tscheligi, J. Fellner   +3 more
openaire   +2 more sources

Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer

International Conference on Learning Representations
Language models can transfer hidden biases during distillation. For example, a teacher that"likes owls"can make its student"like owls"too, even when the training data consists only of lists of numbers.
Simon Schrodi, Elias Kempf, Fazl Barez, Thomas Brox   +3 more
semanticscholar   +1 more source

Subliminal Learning Is Steering Vector Distillation

arXiv.org
Subliminal learning refers to a student language model acquiring a teacher's traits (e.g. a system-prompted preference for owls) when fine-tuned on the teacher's outputs, despite the outputs being semantically unrelated to those traits. It remains poorly
C. Blank, A. Bhatia, Senthooran Rajamanoharan, Arthur Conmy, Neel Nanda   +4 more
semanticscholar   +1 more source

Subliminal Steering: Stronger Encoding of Hidden Signals

arXiv.org
Subliminal learning describes a student language model inheriting a behavioral bias by fine-tuning on seemingly innocuous data generated by a biased teacher model.
George Morgulis, John Hewitt
semanticscholar   +1 more source

The history of subliminal channels

IEEE Journal on Selected Areas in Communications, 1996
In 1978 the United States was considering adopting a national security protocol designed to enable the USSR to verify how many Minuteman missiles the United States had emplaced in a field of 1000 silos without revealing which silos actually contained missiles.
openaire   +1 more source

The Power of the Subliminal: On Subliminal Persuasion and Other Potential Applications

2004
Abstract In the past century, a few hundred articles have been published about subliminal perception and applications such as subliminal persuasion (see Dixon, 1971, 1981). This number suggests a mature field of research, in which, as in any mature field, researchers agree on a number of well-documented findings and argue over more ...
Dijksterhuis, A.J., Aarts, H., Smith, P.K.   +2 more
openaire   +3 more sources

Subliminal Learning is a LoRA Artifact

arXiv.org
Subliminal learning is a phenomenon where language models can transmit behavioral traits to other models through seemingly innocuous data (Cloud et al., 2025). In subliminal learning, a teacher model with a behavioral trait (e.g. obsession with cats) can
Todd Nief, Harvey Yiyun Fu, Mark Muchane, Ari Holtzman   +3 more
semanticscholar   +1 more source

Subliminal wiretapping

Proceedings of the seventh ACM conference on Creativity and cognition, 2009
Subliminal Wiretapping is a subtly interactive artwork that utilizes random number generation modified through mind-matter effects to supply a continuous stream of words. Frequent, personal connections emerge from participants interpreting the stream as the words appear.
openaire   +2 more sources

Learning Through Noise: Why Subliminal Learning Works and When It Fails

arXiv.org
In the context of artificial neural networks, subliminal learning refers to the transfer of task-relevant knowledge or unintended biases from teacher to student models through distillation on task-unrelated input$\unicode{x2013}$output pairs.
Vincent C. Brockers, R. D. Ventzke, Valentin Neuhaus, Belén Hidalgo-Ogalde, V. Priesemann   +4 more
semanticscholar   +1 more source