arXiv Machine Learning

Every Ablation Is a Dose: Counterweights and the Semblance of Self-Repair

The paper investigates the phenomenon of self‑repair in language models, proposing that it arises from a pre‑existing gain in components that act as counterweights when a component is ablated. By modeling interventions as points on a counterfactual axis, the authors derive an affine law for the causal repair response of fine‑grained units, showing that most downstream directions across several models follow this law. They further demonstrate that the slope of this law can be predicted from fixed weights, suggesting that self‑repair is a predictable, counterweight‑driven response rather than a noisy, unexplained effect.

arXiv Machine Learning

Aug 5

Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling

arXiv:2608. 03842v1 Announce Type: cross Abstract: When a language model fails on surface-perturbed input (typos, OCR noise, homophones), "which layer is responsible" has three natural operationalizations: where representations diverge most (sensitivity), where restoring clean activations recovers the prediction (causality), and where a small adapter can repair the damage (compensatory capacity) - and we show these three layer maps dissociate.

By Nathan Labiosa, David Buff, Ena Nayak, Erica Donno

llms computer-vision

arXiv Machine Learning

Aug 13

What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model

arXiv:2608. 10986v1 Announce Type: cross Abstract: A growing class of methods probes a language model by feeding it its own output: self-consistency, iterated refinement, agentic loops.

By Nicol\'as Vera Z\'u\~niga

llms agents

arXiv AI

Jul 3

Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits

arXiv:2607. 01940v1 Announce Type: cross Abstract: Mechanistic interpretability often relies on component-level interventions to discover how a model produces a behavior.

By Zhiren Gong, Zihao Zeng, Chau Yuen, Wei Yang Bryan Lim

llms efficiency benchmarks safety

arXiv AI

Aug 5

A Theory of Conditional Collapse under Low-Rank Weight-Space Ablations: I. The Single-Block Theory and Synthetic Validation

arXiv:2608. 03620v1 Announce Type: cross Abstract: Activation patching and weight-space ablation both claim a component is causally responsible for a behavior, yet they act on different objects: one forward pass versus the parameters behind every forward pass.

By Abdallah Khemais

llms benchmarks

arXiv AI

Aug 11

Decided Upstream, Written Late: Locating and Pricing the Cross-Lingual Refusal Circuit of a Multilingual MoE

arXiv:2608. 08032v1 Announce Type: new Abstract: Safety alignment in multilingual models is uneven: a model that reliably refuses a harmful request in English will often comply with the same request in a lower-resource language.

By Ramakrishna P. Kompella, Aadit Mahajan

safety

arXiv Computation and Language

Sep 25

What a Cross-Model Fixed-Point Census Can and Cannot Arbitrate About Repetition

The paper investigates neural text degeneration by measuring the fixed‑point structure of short‑window argmax maps across 17 pretrained models, using 96 random two‑token starts without prompts. It finds a stable four‑way classification that varies across model families and scales, with some models funneling to a single endpoint token while others do not, and shows that this behavior is not solely determined by training data or corpus frequency. The study demonstrates that repetition phenomena are not uniformly explained by either training data or network architecture alone, highlighting the complexity of neural text generation dynamics.

By Nicol\'as Vera Z\'u\~niga

llms

arXiv AI

Jul 17

Breaking Refusal in the First Half: A Mechanistic Study of the Prefill Jailbreak

arXiv:2607. 14147v1 Announce Type: cross Abstract: Aligned language models refuse harmful requests, but a one-line prefill ("Sure, here is") strips the refusal.

By Alex Kwon

llms safety

arXiv AI

Aug 14

Decomposition of Evidence, Contradiction, and Fragility in Perturbation Responses

arXiv:2608. 12935v1 Announce Type: new Abstract: Perturbation methods explain model decisions by measuring prediction changes under altered inputs, but response magnitude tells us only how much a model reacts, not what that reaction means.

By Lei You

arXiv Machine Learning

Jun 29

The Curse of Multiple Mediators: Hidden Interaction Effects in Activation Patching

arXiv:2606. 27510v1 Announce Type: new Abstract: Activation patching is the primary tool in mechanistic interpretability.

By Sankaran Vaidyanathan, David Arbour, Aaron Mueller, Scott Niekum, David Jensen

llms safety

arXiv AI

Sep 15

Calibrating Interpretability Instruments Before Trusting Their Verdicts

arXiv:2609.14754v1 Announce Type: cross Abstract: Causal claims about large language model (LLM) internals rest on measurements. Those might include a projection, a cosine, an ablation delta, or an i...

By Orion Reblitz-Richardson

llms benchmarks safety

arXiv AI

Jul 28

Gubernaut: A Deterministic Homeostatic Controller for Affect-Regulated LLM Agents, Validated Across Independent Model Families

arXiv:2607. 24339v1 Announce Type: new Abstract: Large language model (LLM) agents inherit reactive failure modes: escalation under provocation, sycophantic drift under flattery, perseveration when stuck.

By Dushyant Sharma

llms agents safety

arXiv AI

Aug 26

Feedback That Backfires: Why Small Language Model Agents Repeat the Call They Just Watched Fail

The study investigates why small language model agents tend to repeat a tool call that just failed. By recording the failed call and its error message in the transcript, the authors measure a negative corrective gain—agents are more likely to repeat the failed action, with a drop of about 1.03 nats per token. The problem is traced to the harness design rather than the model’s understanding of errors, and the authors show that replacing the verbatim call with a runtime-generated description of the failure can reduce this backfiring effect by 76%.

By Esmail Gumaan

llms agents

Related stories