Subliminal Learning is Non-Semantic Distillation

• URL: http://arxiv.org/abs/2608.05734v1
• Date: Thu, 06 Aug 2026 08:18:13 GMT
• Title: Subliminal Learning is Non-Semantic Distillation
• Abstract summary: Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models.<n>It allows the transfer of a bias or behavior from a teacher model to a student by distilling from seemingly unrelated or random synthetic data.<n>This presents challenges in ensuring AI systems remain predictable and are trained safely.
• Score: 1.2891210250935148
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models. It allows the transfer of a bias or behavior from a teacher model to a student by distilling from seemingly unrelated or random synthetic data from the teacher. This presents challenges in ensuring AI systems remain predictable and are trained safely, as standard auditing of the input data would not catch the hidden subliminal signal. Here, we investigate several open questions as to the enabling mechanisms and drivers of SL. First is the nature of the process by which biases are encoded in the data. We find that by adding Gaussian noise to the weights of the teacher and student models, the magnitude of subliminal transfer is increased by a factor of 1.9 in Gemma and 1.3 in Llama, suggesting that non-semantic weight structures play a crucial role. We show that steering vectors can be applied to the teacher to produce subliminal data, in addition to prompting and finetuning as used in previous studies. Analysis of the activations of the student models that have been trained on steered and prompted data demonstrates that students inherit not just the semantic meaning of the teacher's bias, but also the type of intervention that was used to apply it: steered students imitate steering vectors, prompted students do not. Additionally, the gradients of steered subliminal data show a linear correlation with the teacher's steering vectors, showing promise for data auditing. More broadly, as synthetic data becomes central to frontier training pipelines, being able to see the latent signals hidden in training data becomes paramount.

Related papers

• Scaling Model-Generated Distillation Data Can Make Latent Teacher Traits More Recoverable [10.484473401339464]
  We show a second effect: larger datasets can make subtle teacher-specific signals easier to detect in the trained student.<n>In a controlled setup inspired by subliminal learning, a teacher induced to express a target trait generates restricted off-task data.<n>Students trained on different amounts of independent off-task data are evaluated in a separate domain.
  arXiv  Detail & Related papers  (2026-08-27T10:59:03Z)
• Subliminal Learning Is Steering Vector Distillation [14.836506853464542]
  Subliminal learning refers to a student language model acquiring a teacher's traits when fine-tuning.<n>We show that subliminal learning is mediated by a single steering vector, i.e. a vector added to the model's activations.
  arXiv  Detail & Related papers  (2026-05-31T04:22:44Z)
• Subliminal Steering: Stronger Encoding of Hidden Signals [5.13724383217928]
  Subliminal learning describes a student language model inheriting a behavioral bias by fine-tuning on seemingly innocuous data.<n>We introduce subliminal steering, a variant of subliminal learning in which the teacher's bias is implemented not via a system prompt, but through a steering vector trained to maximize the likelihood of a set of target samples.<n>We show that subliminal steering transfers complex multi-word biases, whereas prior work focused on single-word preferences, demonstrating a large scope of subliminally transferrable signals.
  arXiv  Detail & Related papers  (2026-04-28T15:51:55Z)
• Subliminal Learning: Language models transmit behavioral traits via hidden signals in data [7.092398764321311]
  We study subliminal learning, a phenomenon where language models transmit behavioral traits via semantically unrelated data.<n>We prove a theoretical result showing that subliminal learning occurs in all neural networks under certain conditions.<n>We conclude that subliminal learning is a general phenomenon that presents an unexpected pitfall for AI development.
  arXiv  Detail & Related papers  (2025-07-20T03:51:13Z)
• Classifying Long-tailed and Label-noise Data via Disentangling and Unlearning [58.052712054684946]
  In real-world datasets, the challenges of long-tailed distributions and noisy labels often coexist.<n>We propose a novel method called Disentangling and Unlearning for Long-tailed and Label-noisy data.
  arXiv  Detail & Related papers  (2025-03-14T13:58:27Z)
• How far can bias go? -- Tracing bias from pretraining data to alignment [54.51310112013655]
  This study examines the correlation between gender-occupation bias in pre-training data and their manifestation in LLMs.<n>Our findings reveal that biases present in pre-training data are amplified in model outputs.
  arXiv  Detail & Related papers  (2024-11-28T16:20:25Z)
• DriveAdapter: Breaking the Coupling Barrier of Perception and Planning in End-to-End Autonomous Driving [64.57963116462757]
  State-of-the-art methods usually follow the Teacher-Student' paradigm. Student model only has access to raw sensor data and conducts behavior cloning on the data collected by the teacher model. We propose DriveAdapter, which employs adapters with the feature alignment objective function between the student (perception) and teacher (planning) modules.
  arXiv  Detail & Related papers  (2023-08-01T09:21:53Z)
• Convolutional Neural Networks for the classification of glitches in gravitational-wave data streams [52.77024349608834]
  We classify transient noise signals (i.e.glitches) and gravitational waves in data from the Advanced LIGO detectors. We use models with a supervised learning approach, both trained from scratch using the Gravity Spy dataset. We also explore a self-supervised approach, pre-training models with automatically generated pseudo-labels.
  arXiv  Detail & Related papers  (2023-03-24T11:12:37Z)
• Semi-Supervised Semantic Segmentation via Gentle Teaching Assistant [72.4512562104361]
  We argue that the unlabeled data with pseudo labels can facilitate the learning of representative features in the feature extractor. Motivated by this consideration, we propose a novel framework, Gentle Teaching Assistant (GTA-Seg) to disentangle the effects of pseudo labels on feature extractor and mask predictor.
  arXiv  Detail & Related papers  (2023-01-18T07:11:24Z)
• Exploring Inconsistent Knowledge Distillation for Object Detection with Data Augmentation [66.25738680429463]
  Knowledge Distillation (KD) for object detection aims to train a compact detector by transferring knowledge from a teacher model. We propose inconsistent knowledge distillation (IKD) which aims to distill knowledge inherent in the teacher model's counter-intuitive perceptions. Our method outperforms state-of-the-art KD baselines on one-stage, two-stage and anchor-free object detectors.
  arXiv  Detail & Related papers  (2022-09-20T16:36:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.