Transformer circuits show free evolution during SFT, rendering static mechanistic localization inadequate for future parameter updates due to inherent temporal latency.
In-context learning and induction heads.Transformer Circuits Thread
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
Averaging and temporally interpolating text latents in VLAs enables 83% success on novel task combinations in the libero-ood benchmark where SOTA models achieve under 15%.
On Llama-2-7B, path-rerouting magnitude in paired transcoder attribution graphs correlates with jailbreak success (r=0.461), while static node metrics and top-feature ablations do not.
MinMax RNCs are recurrent networks over the min-max semiring that achieve regular language expressivity, log-depth parallel scan, uniformly bounded states, and non-vanishing state gradients while showing competitive empirical performance.
citing papers explorer
-
Navigating by Old Maps: The Pitfalls of Static Mechanistic Localization in LLM Post-Training
Transformer circuits show free evolution during SFT, rendering static mechanistic localization inadequate for future parameter updates due to inherent temporal latency.
-
VLAs are Confined yet Capable of Generalizing to Novel Instructions
Averaging and temporally interpolating text latents in VLAs enables 83% success on novel task combinations in the libero-ood benchmark where SOTA models achieve under 15%.
-
Mechanistic Interpretability of LLM Jailbreaks via Internal Attribution Graphs
On Llama-2-7B, path-rerouting magnitude in paired transcoder attribution graphs correlates with jailbreak success (r=0.461), while static node metrics and top-feature ablations do not.
-
MinMax Recurrent Neural Cascades
MinMax RNCs are recurrent networks over the min-max semiring that achieve regular language expressivity, log-depth parallel scan, uniformly bounded states, and non-vanishing state gradients while showing competitive empirical performance.