Auditability of subliminal learning is constrained by channel location, with initialization-dependent body channels allowing pre-training screens while vocabulary geometry and conditional body channels evade them.
Title resolution pending
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 6roles
background 1polarities
background 1representative citing papers
Transformer Field Theory frames the residual stream as a field, models patching as source insertion, and uses first-order sensitivities plus Green functions to predict and describe responses, with empirical tests on GPT-2 autoregressive models.
Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.
A feedforward graph of heterogeneous frozen LLMs linked by linear projections in a shared latent space outperforms single models on ARC-Challenge, OpenBookQA, and MMLU using just 17.6M trainable parameters.
A learned linear activation bridge achieves high alignment (cosine ~0.97) between Pythia-160M and Pythia-410M states but produces no improvement in downstream multi-hop answering when injected into the receiver.
A-ROM delivers competitive MedMNIST performance via pretrained ViT metric spaces, a concept dictionary, and kNN without backpropagation or fine-tuning, framed as interpretable few-shot learning under the Platonic Representation Hypothesis.
citing papers explorer
-
Channel Location Constrains the Auditability of Subliminal Learning
Auditability of subliminal learning is constrained by channel location, with initialization-dependent body channels allowing pre-training screens while vocabulary geometry and conditional body channels evade them.
-
Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability
Transformer Field Theory frames the residual stream as a field, models patching as source insertion, and uses first-order sensitivities plus Green functions to predict and describe responses, with empirical tests on GPT-2 autoregressive models.
-
Compared to What? Baselines and Metrics for Counterfactual Prompting
Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.
-
Dead Weights, Live Signals: Feedforward Graphs of Frozen Language Models
A feedforward graph of heterogeneous frozen LLMs linked by linear projections in a shared latent space outperforms single models on ARC-Challenge, OpenBookQA, and MMLU using just 17.6M trainable parameters.
-
A Negative Result on Cross-Model Activation Transfer in a Pythia Multi-Hop Setting
A learned linear activation bridge achieves high alignment (cosine ~0.97) between Pythia-160M and Pythia-410M states but produces no improvement in downstream multi-hop answering when injected into the receiver.
-
Toward Aristotelian Medical Representations: Backpropagation-Free Layer-wise Analysis for Interpretable Generalized Metric Learning on MedMNIST
A-ROM delivers competitive MedMNIST performance via pretrained ViT metric spaces, a concept dictionary, and kNN without backpropagation or fine-tuning, framed as interpretable few-shot learning under the Platonic Representation Hypothesis.