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.
Janssen, On a lagrangean for classical field dy- namics and renormalization group calculations of dynam- ical critical properties, Z
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 4roles
background 1polarities
background 1representative citing papers
DYNAMITE is a high-performance solver for dynamical mean-field equations that reaches times up to 10^7 with linear runtime and sublinear memory scaling.
sft-wick is a Python package and formalism that generates diagram tables and computes their integrals for stochastic field theories given an action and observable.
Non-conserved biased tracers debias more rapidly than conserved tracers, leading to time-dependent suppression of large-scale power.
citing papers explorer
-
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.
-
DYNAMITE: A high-performance framework for solving Dynamical Mean-Field Equations
DYNAMITE is a high-performance solver for dynamical mean-field equations that reaches times up to 10^7 with linear runtime and sublinear memory scaling.
-
sft-wick: A formalism and package for Feynman-diagram expansion and evaluation in stochastic field theories
sft-wick is a Python package and formalism that generates diagram tables and computes their integrals for stochastic field theories given an action and observable.
-
Non-conservation and time non-locality of biased tracers
Non-conserved biased tracers debias more rapidly than conserved tracers, leading to time-dependent suppression of large-scale power.