22 / papers

Interpretability

Opening the model up — circuits, features, and what "explaining a behaviour" actually means. 22 open-access papers by Pranay Mahendrakar, each with a permanent DOI.

preprint

A Session Drop Is Not a Signal Ceiling: What Cross-Session Degradation in ECG Biometrics Measures, Why the Same Heartbeat Can Link a Patient Across Years Yet Fail to Unlock a Device, and the Longitudinal Protocol That Would Locate the Loss

Life of Research Electrocardiogram (ECG) biometrics carries two reputations at once. In the authentication literature the heartbeat is a fragile identifier whose accuracy collapses once enrolment and probe come from different sessions; a…

preprint

Harder to Recognise Is Not More Skilled: What a Negative Correlation Between Action-Recognition Accuracy and Rated Nursing Competency Measures, Why the Same Statistic Takes the Opposite Sign in Free Assembly and Expert-Referenced Surgical Scoring, and the Controls That Would Make Recognisability a Competency Signal

Life of Research A 2026 study of egocentric video from 22 nursing students in a pediatric medication-administration simulation reported that a frozen-feature, few-shot action recogniser fitted the sessions of more highly rated students…

preprint

A Smaller Ablation Count Is Not an Easier Model: What Heads-to-Damage Counts Measure When Attention Architectures Differ, Why Grouped-Query Attention, Model Family and Training Cannot Yet Be Separated From Scale, and the Matched Comparisons That Would Say Whether Circuit Analysis Gets Harder With Size

Life of Research A recurring worry in mechanistic interpretability is that circuit analysis becomes harder as language models grow. A 2026 workshop study challenged the worry by comparing Pythia, which uses standard multi-head attention…

preprint

A Right Response Is Not a Learned Law: What Perturbation, Constraint and Interpretability Tests Show About Whether Data-Driven Atmospheric Models Learn Chemistry and Dynamics, Why a Constraint Must Be Expressible Before It Can Be Missed, and the Evidence a Composition Forecast Owes

Life of Research Weather foundation models fine-tuned on reanalysis now issue global air-quality forecasts that match or beat a chemical transport model on most verification targets. A July 2026 study of one of them, Aurora, reports that…

preprint

A Stable Attribution Is Not a Stable Error Rate: What Feature-Attribution Monitoring Can See Under Drift, Why 'More Sensitive Than the Inputs' and 'Silent While Fairness Fails' Are Both Correct, and the Label-Free Statistic the Fairness Alarm Needed

Life of Research Deployed classifiers are often monitored without labels, and feature attributions are one of the signals watched: a production monitoring service alerts when the ranking of SHAP attributions moves, and one line of work…

preprint

Written Is Not Kept: What a Single-Layer Readout of the Residual Stream Licenses When Adjacent Layers Undo Part of Each Other's Writes, Why Cancellation and Iterative Refinement Are Not Rival Findings, and the Persistence Check Interpretability Claims Skip

Life of Research Most feature-based interpretability reads a transformer one layer at a time. A probe, a lens or a sparse dictionary is fitted to the residual stream, or to one layer's output, at a chosen depth, and what it finds there is…

preprint

A Denial Is Also a Report: What First-Person Experience Claims From Language Models Can Show, Why the Mirror Argument Cuts Against Trained Denials as Hard as Against Affirmations, and the Four Links a Self-Report Would Have to Carry

Life of Research When a language model says it is experiencing something, a standard objection holds that the sentence reflects human first-person writing in its training data and is no evidence of anything inside the model; a 2026 paper…

preprint

A Best Match Is Not a Twin: What Feature Similarity Between State Space Models and Transformers Certifies, Why the Null and the Shared Inputs Decide the Verdict, and the Controls That Would Show Two Architectures Learned the Same Thing

Life of Research Selective state space models such as Mamba compress the past into a fixed-size state, while Transformers attend over the whole context, and a growing set of studies asks whether the two learn the same internal features. A…

preprint

Newly Important Is Not Newly Built: What Reasoning Post-Training Adds Inside a Language Model, Why the Evidence Separates Representation From Wiring From Deployment, and the Cross-Model Tests That Would Show an Attention Head Is New

Life of Research Reasoning models are built by post-training a pretrained model with supervised fine-tuning on teacher traces, with reinforcement learning against a verifier, or with both, and a running dispute asks whether this builds new…

preprint

Which Way the Verdict Falls: Noise-Induced Flips in an LLM Bias Judge Run Toward the Label Its Prompt Makes Easiest to Reach, the Harm-Detection Literature Records the Opposite Sign, and Whether Noise Inflates a Bias Score Depends on the Score

Life of Research Language models are now used as judges that read a passage and decide whether it leans toward a stereotype, and the passages they read are often noisy. A 2026 study corrupted 3,822 stereotype-probing responses with typos…

preprint

Hard to Read Is Not Hard to Audit: Why Degrading a Text Image Raises Jailbreak Success in One 2026 Study and Lowers It in Another, Four Accounts That Fit the Same Curve, and the Ablations That Would Separate Them

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. Rendering text as an image is now proposed as a way to compress long contexts for multimodal language models, and a 2026 study reports that the same move opens a…

preprint

Which Signal Did the Synthesis Import? Information, Re-presentation and Selection in Synthetic Pretraining, Why 'the Verifier Is Doing the Work' Holds Only Where Correctness Is the Target, and the Controls That Would Assign the Credit

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. The expectation that language models will keep improving after the supply of human text runs out rests on a claim about synthetic data: that a model, or a family of…

preprint

Repair Without a Repairer: Five Mechanisms Called Self-Repair in Neural Networks, Why Much of the Measured Restoration Is Restored Scale, and What Would License the Claim That a Network Restores Its Own Performance

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. "Self-repair" is used in at least three research communities that do not cite one another. In transformer interpretability it names the Hydra effect: when one…

preprint

Circuits Are Estimates, Not Mechanisms: A Recovered Subgraph Is One Draw From a Class Set by the Ablation, the Dataset and the Grain, the Results That Defend Circuits Hold at the Level of That Class, and What Ablation Evidence Licenses at Each Level

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. A circuit in a language model is usually presented as the mechanism by which the model performs a task: a sparse subgraph, found by ablation or patching, that…

preprint

Dictionaries Without Ontologies: What a Sparse Autoencoder Latent Is Evidence For, Why Decodability, Use and Reproducibility Do Not Add Up to Canonicity, and the Identifiability Premise No Real-Model Result Has Checked

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. Sparse autoencoders are now the default instrument for decomposing language model activations, and their latents are routinely reported as the model's features: units…

preprint

Agreement Is Not a Criterion: What Matching Human Labels Certifies About an LLM Judge, Why a Shared Shortcut Passes the Test, and Why Every External Criterion Found So Far Validates a Use Rather Than a Judge

Life of Research (c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. An automatic judge is usually accepted on one piece of evidence: it agrees with human preference labels about as often as humans agree with each other. That number is…

About this research line

What has Pranay Mahendrakar published on Interpretability?

Pranay Mahendrakar has published 22 open-access papers on Interpretability: "A Session Drop Is Not a Signal Ceiling", "Harder to Recognise Is Not More Skilled", "A Smaller Ablation Count Is Not an Easier Model", "A Right Response Is Not a Learned Law", "A Stable Attribution Is Not a Stable Error Rate", "Written Is Not Kept", "A Denial Is Also a Report", "A Best Match Is Not a Twin", "Newly Important Is Not Newly Built", "Which Way the Verdict Falls", "Hard to Read Is Not Hard to Audit", "Wrong Together Is an Item-Level Quantity", "Which Signal Did the Synthesis Import? Information, Re-presentation and Selection in Synthetic Pretraining, Why 'the Verifier Is Doing the Work' Holds Only Where Correctness Is the Target, and the Controls That Would Assign the Credit", "Repair Without a Repairer", "Circuits Are Estimates, Not Mechanisms", "Dictionaries Without Ontologies", "Agreement Is Not a Criterion", "Predicting Your Own Failure Is Not Predicting Another Model's Advantage", "The Self-Verification Gap", "Memory Architectures Beyond Attention", "Mechanistic Interpretability of In-Context Learning" and "Sheaf-Theoretic Semantics and Quantum Contextuality in Large Language Models". Each is deposited on Zenodo with a permanent DOI and is free to read.

← All research topics