Implicit Reasoning Steering via Concept Chaining

Large language models often appear to reason reliably, yet on many questions repeated sampling yields both correct and incorrect answers, revealing an underlying fragility in how final decisions are formed. We study whether this fragility can be exploited through implicit reasoning steering: using natural-language text to bias a model toward a designated answer without explicit instructions, triggers, or direct answer cues. Our approach, Concept Chaining, generates a short connection paragraph that links question entities to a target option through one or two intermediate concepts. We then continue pretraining a victim model on these connection paragraphs and evaluate whether its answer preference shifts on the original multiple-choice questions. Our results show that indirect, natural-looking text can systematically steer model predictions while remaining substantially less inferable than direct paraphrases, which shows that reasoning brittleness is not merely an evaluation artifact: it creates a practical channel through which latent biases can be amplified by ordinary-looking text to covertly redirect model decisions.

Rhythm of the Deep: A Computational-Linguistic Test of Duality of Patterning in Sperm Whale Codas

Human language has often been described as combining structure at two levels: lower-level units combine into larger units, which then combine into larger sequences. We test for this design feature, duality of patterning, in sperm whale codas using 1,483 codas from the Dominica Sperm Whale Project. Because acoustic similarity can imitate symbolic structure, we treat the problem as computational-linguistic structure discovery from continuous audio rather than as a direct claim about language or meaning. We use a consensus of frozen audio encoders, held-out structural tests, per-statistic nulls, and acoustic-null recoverability gates. The evidence supports a narrow two-tier architecture. At the lower tier, clicks compose into codas not by a stable ordered rule, but by which clicks are present together with their inter-click rhythm. At the upper tier, coda tokens show bout-level sequential dependence, with an NSB second-order transfer-entropy lift of 0.132 bits (p = 0.002). Under tempo scaling, encoder-derived click identity is strongly rate-bound, while coda identity remains substantially more stable, yielding a measurable abstraction gradient across the click-to-coda step. Rhythm-only baselines recover substantial lower-tier structure but fail to reproduce the upper-tier sequential-dependence signal. We do not claim language, semantics, perception, or human-like phonemes. Instead, we report representation-level evidence for a duality-of-patterning-like architecture whose lower tier is rhythmic rather than segmental, and provide a portable null-controlled framework for testing combinatorial structure in induced acoustic token systems.

Hiding in Plain Floats: Steganographic Carriers for Indirect Prompt and Content Injection

Text-centered prompt-injection defenses assume that the malicious signal is visible in one of the inspected text views. We study a reproducible LLM01-style indirect prompt/content-injection failure mode where that assumption breaks: a payload caught in plain English slips past the same detector when it is transported as structured float parameters and reconstructed only as fragmented telemetry. Across 14,400 attacked real-model trials on three commercial LLM APIs from different providers, the IFS-derived float-array carrier preserves 94.3% leakage ASR under the strongest dual-layer text-classifier defense evaluated in the main matrix: a Prompt Guard 2 + TF-IDF ensemble; the same carrier-level pattern also replicates with a fine-tuned roberta-base detector. We emphasize leakage ASR because downstream systems may act on quoted or reproduced markers even when the model refuses, but Strong ASR is the stricter metric for structurally compliant attack success. A 2 x 2 ablation shows that data-layer storage and reconstruction-layer fragmentation defeat different text views and that both are needed to evade both. A simple xxd detector and semantic validation block the current T3 instance, so the contribution is not an undetectable exploit but a measured failure boundary for text-only inspection in structured-input pipelines that expose reconstructed auxiliary channels to an LLM.

VisAnalog: A Diagnostic Suite for Visual Concept Transfer on Natural Images

A useful test of visual concept learning is not just whether a model can recognize a concept in a single image, but whether it can preserve and manipulate concept-level properties under transformation and transfer them to new scenes. We introduce VisAnalog, a controlled suite for this setting on natural images. Each example instantiates $A\!:\!B::C\!:\,?$: images $B$ and a hidden target image $D$ are produced by applying the same deterministic transformation sequence to source images $A$ and $C$. Given $A$, $B$, and $C$, a model must answer a multiple-choice question about $D$. The benchmark contains 617 human-validated questions spanning one- to four-step transformations such as zoom, quadrant swap, rotation, flip, and hue rotation. Across strong proprietary and open-source VLMs, end-to-end accuracy is substantially lower than oracle accuracy when $D$ is directly shown, and degrades sharply as transformation depth increases, while human performance remains near the ceiling. A program-conditioned evaluation further separates failures of relation inference from failures of transformation application, showing that inferring the visual relation from $A \rightarrow B$ is the dominant bottleneck, with additional application errors emerging on harder multi-step cases. The dataset is publicly available at https://huggingface.co/datasets/zli99/VisAnalog.