arXiv subjects

John Winnicki

Publications and source records attributed to John Winnicki.

7 recordsLinked to original sources

arXiv2026

How Do Language Models Choose Between Context and Memory?

When contextual information conflicts with the knowledge stored in model parameters, activation directions can be used to decode and steer which source the model follows. However, steering along a direction does not establish causality: whether the unedited model would naturally use that direction or whether the direction is reusable across tasks. We test these distinctions through counterfactual experiments in unambiguous settings. First, we estimate authority directions from agreement prompts, in which the context and parametric knowledge support the same answer. We then interchange naturally occurring coordinates along these directions between matched prompts that direct the model to prioritize either the supplied context or its parametric knowledge. Across Qwen, Llama, and OLMo models, this intervention reproduces 30-68% of the authority-induced shift in source choice, whereas matched controls reproduce almost none. To test cross-task reuse, we learn authority directions on two tasks separately and see that cross-task transferability closes only 9% of the authority gap while the local direction learned on the given task closes 57%. These results distinguish authority representation, causal use, and cross-task causal reuse, and suggest that authority computations may be task-dependent, rather than reusable across tasks.

cs.LG↗

arXiv2026

When Does Activation Steering Change What a Model Computes From?

Activation steering can reliably change an agent's output by modifying its internal activations. Yet arriving at the same answer need not involve the same computation: behavioral equivalence does not imply mechanistic equivalence. We test whether an activation edit changes the state used in subsequent computation or biases that computation toward the desired output. In a controlled state-tracking task, trace supervision creates an editable register: changing its internal value makes the model apply the next operation to the edited state. We next ask whether widely used mean activation steering admits a similar interpretation: does steering recreate the internal configuration the model naturally uses when performing a task? If so, transplanting that activation should be effective, and steering should remain effective at the scale of the natural source-to-target activation change. Neither prediction holds in the two selected model-task settings. Replacing the activation at one layer produces less than 2% of the target-answer margin gain from patching through all remaining layers, while natural-scale steering is similarly ineffective. The steering vectors have norms 77 and 26 times the median natural change in Qwen and Llama, producing 54% and 82% of the reference effect. Thus, a successful steering intervention need not reproduce the natural target activation at the intervention layer. More generally, an intervention should be interpreted as changing the computational state only when a later computation uses the edited value according to the semantics of that state.

cs.LG↗

arXiv2026

Domain-Filtered Knowledge Graphs from Sparse Autoencoder Features

Sparse autoencoders (SAEs) extract millions of interpretable features from a language model, but flat feature inventories aren't very useful on their own. Domain concepts get mixed with generic and weakly grounded features, while related ideas are scattered across many units, and there's no way to understand relationships between features. We address this by first constructing a strict domain-specific concept universe from a large SAE inventory using contrastive activations and a multi-stage filtering process. Next, we build two aligned graph views on the filtered set: a co-occurrence graph for corpus-level conceptual structure, organized at multiple levels of granularity, and a transcoder-based mechanism graph that links source-layer and target-layer features through sparse latent pathways. Automated edge labeling then turns these graph views into readable knowledge graphs rather than unlabeled layouts. In a case study on a biology textbook, these graphs recover coherent chapter and subchapter-level structure, reveal concepts that bridge neighboring topics, and transform messy sentence-level activity containing thousands of features into compact, readable views that illustrate the model's local activity. Taken together, this reframes a flat SAE inventory as an internal knowledge graph that converts feature-level interpretability into a global map of model knowledge and enables audits of reasoning faithfulness.

cs.AI↗

arXiv2026

Flash-SD-KDE: Accelerating SD-KDE with Tensor Cores

Score-debiased kernel density estimation (SD-KDE) achieves improved asymptotic convergence rates over classical KDE, but its use of an empirical score has made it significantly slower in practice. We show that by re-ordering the SD-KDE computation to expose matrix-multiplication structure, Tensor Cores can be used to accelerate the GPU implementation. On a 32k-sample 16-dimensional problem, our approach runs up to $47\times$ faster than a strong SD-KDE GPU baseline and $3{,}300\times$ faster than scikit-learn's KDE. On a larger 1M-sample 16-dimensional task evaluated on 131k queries, Flash-SD-KDE completes in $2.3$ s on a single GPU, making score-debiased density estimation practical at previously infeasible scales.

cs.DC↗

arXiv2026

Allocate Marginal Reviews to Borderline Papers Using LLM Comparative Ranking

This paper argues that large ML conferences should allocate marginal review capacity primarily to papers near the acceptance boundary, rather than spreading extra reviews via random or affinity-driven heuristics. We propose using LLM-based comparative ranking (via pairwise comparisons and a Bradley--Terry model) to identify a borderline band \emph{before} human reviewing and to allocate \emph{marginal} reviewer capacity at assignment time. Concretely, given a venue-specific minimum review target (e.g., 3 or 4), we use this signal to decide which papers receive one additional review (e.g., a 4th or 5th), without conditioning on any human reviews and without using LLM outputs for accept/reject. We provide a simple expected-impact calculation in terms of (i) the overlap between the predicted and true borderline sets ($ρ$) and (ii) the incremental value of an extra review near the boundary ($Δ$), and we provide retrospective proxies to estimate these quantities.

cs.DL↗

arXiv2025

LLMs are Overconfident: Evaluating Confidence Interval Calibration with FermiEval

Large language models (LLMs) excel at numerical estimation but struggle to correctly quantify uncertainty. We study how well LLMs construct confidence intervals around their own answers and find that they are systematically overconfident. To evaluate this behavior, we introduce FermiEval, a benchmark of Fermi-style estimation questions with a rigorous scoring rule for confidence interval coverage and sharpness. Across several modern models, nominal 99\% intervals cover the true answer only 65\% of the time on average. With a conformal prediction based approach that adjusts the intervals, we obtain accurate 99\% observed coverage, and the Winkler interval score decreases by 54\%. We also propose direct log-probability elicitation and quantile adjustment methods, which further reduce overconfidence at high confidence levels. Finally, we develop a perception-tunnel theory explaining why LLMs exhibit overconfidence: when reasoning under uncertainty, they act as if sampling from a truncated region of their inferred distribution, neglecting its tails.

stat.ME↗

arXiv2025

SD-KDE: Score-Debiased Kernel Density Estimation

We propose a novel method for density estimation that leverages an estimated score function to debias kernel density estimation (SD-KDE). In our approach, each data point is adjusted by taking a single step along the score function with a specific choice of step size, followed by standard KDE with a modified bandwidth. The step size and modified bandwidth are chosen to remove the leading order bias in the KDE. Our experiments on synthetic tasks in 1D, 2D and on MNIST, demonstrate that our proposed SD-KDE method significantly reduces the mean integrated squared error compared to the standard Silverman KDE, even with noisy estimates in the score function. These results underscore the potential of integrating score-based corrections into nonparametric density estimation.

cs.LG↗