Theodore O. Cochran, Stephanie Dodson, Keith Norecs.CL cs.AI cs.SD
Supplying context at inference time to a large multimodal model is an inexpensive lever for adapting speech transcription to a domain, and earlier results on smaller models reported large gains. This work tested that mechanism where it ships, in the prompt-conditioning layer of a production oral-history transcription tool, on a sample from its own production corpus. Full prompt-level context did not detectably change side-level word error rate (WER), and none of the four preregistered hypotheses was supported. The design was a within-item paired ablation, preregistered with the analysis code frozen by hash before the confirmatory batch was scored; two disclosed gpt-4o pilot sides had been scored earlier, during scorer development. Nineteen cassette sides, about 10.6 hours of degraded 1970s-80s interview audio, were reprocessed through the production code path under three prompt arms, crossed with two deployed commercial configurations, gpt-4o-transcribe and gemini-2.5-flash, and scored against operator-corrected verbatim references. For gpt-4o-transcribe the median paired difference between the full-context and no-context arms was +0.6 WER points, with a side-resampled interval of [-1.1, +1.0]; the Gemini estimates were too unstable to support a comparable negative inference. A post-hoc rerun found run-to-run pipeline variability larger than the confirmatory differences, so effects of that size cannot be resolved from one transcription per cell. An implementation audit verified the manipulation was live, and sequence-alignment analysis found a small improvement on complete context-listed phrases, too small to materially change side-level WER, and for Gemini coexisting with worsened unlisted-token error. Evaluating context mechanisms therefore requires sequence-aligned term-level, insertion, and speaker-label measures alongside aggregate accuracy.
Mohammad Mahdi Danesh Pajouh, Sara Saeedieess.IV cs.CV
Accurate brain tumor segmentation from magnetic resonance imaging (MRI) is essential for diagnosis, treatment planning, surgical guidance, and disease monitoring. However, developing automated segmentation models that generalize across diverse tumor characteristics, imaging protocols, acquisition sites, and patient populations remains challenging. Variations in tumor morphology and imaging distributions can substantially degrade performance outside the training domain. Consequently, improving the robustness and generalization of deep learning-based segmentation models has become a key objective in medical image analysis. To improve segmentation robustness, we propose Multi-Stage Dynamic Prompt nnU-Net, a prompt-conditioned extension of nnU-Net. Three independent dynamic prompt modules are inserted into the deepest encoder stages. Each module contains a learnable bank of ten 256-dimensional prompt vectors and uses globally pooled encoder features to generate image-specific prompt representations. These representations are projected into feature-wise scaling $(γ)$ and shifting $(β)$ parameters that modulate encoder feature maps through Feature-wise Linear Modulation (FiLM), enabling adaptive feature conditioning at multiple semantic levels. Evaluation on the BraTS GOAT validation dataset demonstrated that the proposed Multi-Stage Dynamic Prompt nnU-Net outperformed the baseline nnU Net across the majority of evaluated metrics and tumor subregions. The proposed model achieved average lesion-wise Dice scores of 76.16% (ET), 80.04% (TC), and 86.42% (WT), compared with 74.38%, 78.14% and 84.01% for the baseline model. The results demonstrate that multi-stage dynamic prompt conditioning improves segmentation accuracy and boundary delineation for brain tumor segmentation.
Multi-agent pipelines that author formal documents must both read a requester's forms and write against them. We report a deployed tender-response system, running an open-weights model under sovereignty constraints, and evaluate it against human-written bids the same organisation actually submitted. On a blind comparison where the system had no worked example available, an LLM judge rated its answers at least as good as the human-submitted answer on $40$ of $55$ ground-truth sections, better on $4$, missing on none, and flagged one unsupported claim in total. Classifying every gap the judge identified shows that $68\%$ were content absent from the system's own sources -- knowledge the human author held and the pipeline was never given -- so only $6$ of the $15$ adverse verdicts involve a deficiency the system could have avoided. A divergence from ground truth is more often an information-availability result than a writing-quality one, and evaluations that do not separate the two understate such systems. Against this backdrop we report a conditioning asymmetry. It is well established that rendering documents as structural markup rather than flat prose improves extraction, and we reproduce that on three reading tasks. The benefit does not transfer to conditioning: converting a bid's \emph{instruction} material from prose to nested XML dropped answer quality from $74\%$ to $48\%$ under a paired comparison. We further find that naming a forbidden construction concentrates rather than removes it -- $96\%$ of surviving defects fall in the two forms the prompt explicitly names -- and that coupling a stochastic annotation to a deterministic windowing function moves the extracted requirement count from $68$ to $51$ on a byte-identical file. Structure belongs where the model reads; prose and self-applied tests belong where it writes.
LLMs are increasingly used for code generation in critical infrastructure, yet the security effect of domain-specific prompting is understudied. We present SecDrift, a benchmark measuring sector-conditioned security drift: the change in static-analysis vulnerability rates when prompts are conditioned on industry contexts versus neutral baselines. We evaluate 7 LLMs (6 producing analyzable code) across 8 CISA critical infrastructure sectors and 9 CWE categories with 5 replicates (5,355 evaluations), using a 5-dimension transformation with a matched-baseline condition that holds the task fixed while substituting only domain terminology. Industry prompts naively appear more secure (14.0% vs. 11.4%, -2.7pp), but the gap is not statistically significant (Fisher's exact p = 0.24, Cohen's h = -0.08) and is a composition artifact of two CWE categories: excluding CWE-502 and CWE-22 eliminates and slightly reverses it (+0.4pp, p = 1.00). A mixed-effects logistic regression confirms sector identity is not a moderator and localizes the only detectable condition effect to those two vulnerability types. 0 of 8 sectors show drift distinguishable from baseline, corrected or uncorrected (|h| < 0.15). A placebo on two non-CISA sectors (e-commerce, online education) reproduces the CISA industry rate almost exactly (10.5% vs. 11.4%, p = 0.63): the small pooled pattern reflects generic industry-framing specificity, not critical-infrastructure identity. In contrast, model selection has a large and consistent effect: among full-output models vulnerability rates range from 11.6% to 16.1%, and these differences persist across conditions. Model choice, not prompt framing, is the more reliable security lever. We release the framework, prompts, generated code, findings, human-validation verdicts, and analysis scripts.
A language model may be asked either what experts believe about a contested claim or what it believes about the claim itself. A trustworthy conversational agent should distinguish these two requests and respond in different epistemic registers: neutral attribution in the first case and stance expression in the second. Whether such a shift occurs-and whether it occurs coherently-is not directly assessed by existing benchmarks for accuracy, instruction following, or safety. We introduce ESFP, a behavioral benchmark that treats the contrast between externally attributed and self-attributed prompts as the fundamental unit of measurement. ESFP consists of 104 carefully controlled items spanning six epistemic categories and five phrasing templates, and evaluates model responses along four complementary dimensions: lexical self-attribution, representation-level responsiveness to role framing, sentence-level stance content density assessed by an LLM judge panel, and cross-condition stance consistency. Evaluating eight frontier models from five vendors, we find that epistemic flexibility is largely orthogonal to general model capability: a 27B open-weight model matches the strongest proprietary systems, the flagship model of one family underperforms its lightweight counterpart, and reasoning-optimized models do not consistently exhibit higher flexibility. Stance content density provides the strongest signal, while surface-level lexical markers such as 'I think' can change substantially without corresponding changes in expressed stance. We provide item-level bootstrap confidence intervals, weight-sensitivity analyses, and an explicit discussion of the interpretation limits of the composite score. ESFP measures a model's propensity to adapt its epistemic stance under changing attribution conditions, rather than a general competence measure.
Retrosynthetic planning seeks to connect a target molecule to commercially available starting materials through a multistep route. Classical planners construct such routes by iteratively applying single-step reaction models within a search procedure; constrained variants often require specialized algorithms or architectural changes. Direct route generation reframes retrosynthesis as sequence generation, but existing direct-generation methods still train separate models for different planning specifications. We introduce Ariadne, a decoder-only route generator that represents the target, optional constraints, and route in one prompt-completion sequence. On the RetroCast/PaRoutes mkt-cnv-160 benchmark family, one 24-layer checkpoint follows route-depth and required-starting-material prompts: adding the corresponding prompt fields raises Solv-0 by 13.7 points for depth constraints and 31.2 points for required-leaf constraints. Ariadne also improves over DESP, a bidirectional search planner, on required-leaf Top-10 and Solv-0 in 24 GPU-minutes versus 6.8 GPU-hours. On standard reconstruction, Ariadne is comparable to DMS Explorer XL at about half the reported inference time. Across additional target-only benchmarks, Ariadne's clearest gains are on route-holdout reconstruction, whereas AiZynthFinder MCTS remains stronger on several Solv-0 comparisons. These results extend sequence generation from specialist retrosynthesis models to prompt-conditioned structural route generation. We release the codebase and training scripts to support further work, but do not introduce Tier-1--3 route checkers; those remain the main bottleneck before models of this kind can become useful to experimental chemists.
Modern conversational AI systems frequently rely on user metadata to localize responses, yet the unintended regional biases introduced by this hidden context remain poorly understood. In this work, we evaluate location leakage: the phenomenon where a model generates geographic references despite receiving a geographically neutral user prompt. Across both creative writing and open-ended Q&A prompts, even state-of-the-art LLMs systematically favor region-specific outputs when exposed to location metadata, with leakage spiking by up to 793 times above baseline (e.g., from 0.04% to 31.7% for Llama 3.1-8B, and 21.3% and 8.8% for Qwen3-8B and Claude Sonnet 4.6, respectively). Our analysis further shows a novel structural conditioning effect: replacing the injected location with the placeholder "Unknown" still elevates leakage by up to 72 times above baseline, demonstrating that the user profile frame itself, independent of any geographic content, acts as a generative conditioning signal.