overfeed.news
Anuncio

NexTalk

Automatiza la primera respuesta. Mantén el toque humano.

Chat en vivo con agentes de IA en tu modelo y tu clave de API, con traspaso a tu equipo.

Empezar gratis

Tema

LLMs

2972documentos

226últimos 7 días

arXiv AI Papers

Discovering Models for Experimental Quantum Systems with Agentic AI

A central obstacle in realizing color center cavity-QED-based quantum technologies is inhomogeneity, or disorder, among individual emitters. As the number of emitters increases, so does the complexity of the experimental data, making it difficult to quantify the present physical mechanisms of disorder. Here we introduce an artificial intelligence-driven model discovery framework in which a large language model (LLM)-based agent, equipped with tools and a structured workflow, proposes candidate Hamiltonian and Lindbladian models, generates computational tools, and quantitatively evaluates competing physical hypotheses. We apply the workflow to three previously published experimental cavity-QED datasets collected from color centers in diamond and silicon carbide. Across the experiments, the agent successfully identifies physical models for the system and characterizes the disorder. In assessing reproducibility, we find that run-to-run variability is concentrated primarily in the research plan generation step, while the computational analysis itself is substantially more consistent. Our results suggest that structured LLM agent workflows can support model discovery from quantum optical experimental data while simultaneously allowing for human expert supervision and collaboration.

en
arXiv AI Papers

Enhancing Language Model Reasoning with Structured Multi-Level Modeling

Inference-time scaling enhances a model's reasoning by extending its chain-of-thought (CoT). However, existing approaches typically rely on a single policy trained with outcome-reward reinforcement learning (RL), which often suffers from long-horizon plan failures where the implicit plan drifts from valid strategies, especially for small LMs with limited capacity. To address this, we propose Multi-Level Reasoning (MLR), which reformulates long-CoT generation as a two-level stochastic process. A high-level planner generates structured step descriptors specifying both the reasoning mode and the semantic subgoal. The low-level executor then produces detailed reasoning conditioned on these descriptors, forming an alternating plan--execute loop. To maintain scalability, we adopt a minimal design where the base model serves as the low-level policy and a lightweight LoRA module implements the high-level policy. For training, we observe that outcome-reward RL provides sparse and delayed feedback for long trajectories (e.g., several thousand tokens), hindering credit assignment. We therefore introduce iterative Step-DPO, a process-level preference optimization scheme that leverages Twisted Sequential Monte Carlo (TSMC) to provide scalable stepwise supervision. This yields more effective training, improved stability, and higher accuracy. Extensive experiments on challenging math, science, and logical reasoning benchmarks show that, under the same reduced data budget (10% SFT and 5% preference relative to the DeepSeek-R1 distillation setup), MLR outperforms both SFT-based distillation and strong RL/preference-optimization baselines across multiple base models and tasks. Moreover, MLR exhibits slower performance degradation on long-horizon reasoning, demonstrating stronger robustness under extended CoT generation.

en
arXiv AI Papers

Trail: Accurate Local Smart Home Control Using Decision Models

Natural-language driven smart home control remains challenging: cloud-hosted Large Language Models (LLMs) raise privacy concerns, while local Small Language Models (SLMs) can generate invalid actions or misinterpret requests. We introduce Trail (Typed Registry-Anchored Inference for Language commands in Smart Homes), a fully local framework that replaces open-ended command generation with hierarchical typed decisions. Trail exploits the natural structure of a home's action space, selecting a room, device, and operation from candidates derived from the runtime device registry. This ensures valid actions while enabling confidence-based abstention when requests cannot be reliably resolved. We also design a structured memory module that uses typed decisions to support user preferences, device references, and multi-turn interactions. To evaluate Trail, we extend HomeBench and generate three datasets: Trek-Base, which tests natural-language variation; Trek-Robust, which evaluates robustness to changes in the home and its description; and Trek-Memory, which assesses conversational memory across multi-turn interactions. Trail achieves 85.3% accuracy on Trek-Robust versus 69.8% for a finetuned Qwen3.5-4B baseline, while remaining competitive on Trek-Base requests. With memory, Trail resolves 75.6% of the requests that depend on earlier turns, compared with 42.5% for the finetuned baseline given conversation history.

en
arXiv AI Papers

Words Before Pixels: Selective Modality Routing for Vision-Language Model Unlearning

Vision-language model (VLM) unlearning is often treated as an objective-design problem: given multimodal forget data, the goal is to optimize a loss that removes undesirable behavior while preserving retained capabilities. We argue that this view overlooks an equally important data-centric question: how should each forget example be represented before optimization? Existing methods typically keep image-conditioned failures in their original image-text form, implicitly assuming that pixels are the right route for unlearning whenever pixels appear in the input. We challenge this assumption by constructing modality-decomposed versions of VLM benchmarks, enabling controlled comparisons among image-only, text-only, and multimodal representations of the same examples. Our modality-mixing experiments show that unsafe instruction-following behavior is more effectively suppressed through textual representations than through images alone, while purely text-only unlearning can still weaken visual grounding when the target behavior depends on visual evidence. Motivated by this trade-off, we introduce ModRoute, a gradient-guided per-sample modality routing strategy that selectively converts high visual-pressure examples to text-only form while keeping the remaining examples multimodal. Notably, on VLGuard with LLaVA-1.5-7B, ModRoute has the strongest composite unlearning-utility score of 0.847 at γ=0.5, a 2.5% improvement over the best random-switching score and 46% better than pure multimodal or text unlearning. Overall, our results show that effective VLM unlearning depends not only on the objective, but also on the composition and per-sample modality routing of the unlearning data.

en
arXiv AI Papers

Conversational Interaction for LLM-Based Clinical Text-to-SQL: A User Study with Healthcare Professionals

Conversational text-to-SQL interfaces could give healthcare professionals natural-language access to electronic health record (EHR) data, but evidence for their real-world usability remains limited: prior evaluations have relied largely on curated benchmarks with fully specified questions or on simulated users. We report a user study in which 14 physicians who regularly work with EHR data used both a single-turn and a conversational multi-turn text-to-SQL interface over the MIMIC-IV database in a within-subjects design; dialogue analyses additionally drew on interaction logs from 21 healthcare professionals across the main study and two pilot studies. The conversational interface was consistently favored: mean System Usability Scale scores rose from 68.2, near the industry average, to 85.9 (p<0.001), with significant advantages across technology-acceptance constructs including performance expectancy, effort expectancy, self-efficacy, and behavioral intention. In post-study forced-choice comparisons, 93% of participants reported greater overall satisfaction with the conversational interface, 86% preferred it for real-world adoption, and 71% trusted its results more; no participant preferred or trusted the single-turn interface more. Turn-level analysis of 425 logged user turns showed that freely composed queries triggered clarification requests in 24.8% of cases versus 1.8% for predefined, fully specified questions, and 64.3% of clarification episodes stemmed from EHR- or clinical-domain knowledge. Conversational interaction is a practical prerequisite, rather than a convenience, for natural-language EHR access. The frequent need for clarification in realistic clinical queries further underscores the importance of conversational interfaces, while benchmark-style evaluation can substantially underrepresent the ambiguity of real clinical use.

en
arXiv AI Papers

Multi-Estimate Elicitation Does Not Uniformly Debias Language Model Judgments

Does eliciting five weighted estimates within a single response change how a language model responds to information? Across anchoring, hindsight, base-rate, conjunction, and status-quo/default tasks, we compare direct numerical answers with a multi-estimate response format, contrasting source framings, explicitly random cues, and probability-consistency constraints. We evaluate three completed configurations--gpt-6.1-sol, claude-sonnet-5.5, and Qwen3-8B (reasoning disabled, analyzed descriptively). In the first two configurations, direct elicitation yielded near-zero irrelevant-anchor indices and status-quo effects, while remaining responsive to informative sources and informed advice. Of 23 preregistered confirmatory tests, six rejected after Holm adjustment; however, all six absolute point estimates fell below their prespecified smallest effect sizes of interest without establishing equivalence. Moreover, while separate versus joint elicitation shifted conjunction judgments, interpretation is constrained by between-call noise. Ultimately, multi-estimate elicitation failed to produce uniform debiasing or response attenuation across tasks and models, though post hoc correlations between candidate spread and anchoring error motivate independent predictive validation. Materials: https://github.com/asj252/llm-judgment-studies

en
arXiv AI Papers

Do Orthogonality Constraints Help Language-Model Training Beyond Sphere Optimization?

A growing body of work reports benefits from orthogonality constraints in fine-tuning large language models (LLMs). These results raise an open question: do orthogonality constraints also help LLM pre-training? We conduct a systematic empirical study of decoder-only LLM pre-training with models ranging from 360M to 3.1B parameters, trained on nearly 200B tokens. We examine fourteen variants of orthogonality training: some constrain the weight matrix itself, while others use orthogonal factors to reparameterize it. Because a strictly orthogonal square matrix loses half of its degrees of freedom, we also study block-wise orthogonality, where every d rows form a row-orthonormal (Stiefel) matrix and the block size d is a hyperparameter. Among block sizes ranging from 1 to the shorter side of the matrix, no d>1 used throughout the network reliably outperforms d=1, and large blocks are clearly worse; the one placement that helps, symmetric positive-definite factors on the query and key projections, loses its small gain as models grow. At d=1, the constraint reduces to sphere optimization, with each row direction on its own unit sphere, which by itself already improves on tuned Muon. The same two findings hold for a diffusion transformer trained on ImageNet, and after fine-tuning, downstream accuracy does not separate the parameterizations. Our conclusion is that, across the realizations we tried, orthogonality between rows does not help language-model pre-training beyond sphere optimization. Since this appears to conflict with several prior results, we re-examine five prior methods in pre-training and fine-tuning that report orthogonality improvements. In each case the reported gain is attributable to numerical precision, training horizon, optimizer realization, or per-row norm control rather than to the orthogonality the methods impose.

en
arXiv AI Papers

Saidzoda-Tajik-E5-Large: Fine-Tuning a 560M Embedder for Tajik Retrieval

Finding the passage that answers a question is the first step of most practical systems built on documents, and for Tajik we found no openly described model for it. We present Saidzoda-Tajik-E5-Large, a Tajik text embedder built from multilingual-e5-large-instruct (560M parameters). We asked our own Tajik language model to write questions for passages from Tajik laws, tax documents, Wikipedia and web text, and trained on 235,140 question and passage pairs for one epoch, with hard negatives chosen by another embedder and English and Russian pairs mixed in so the model would not forget those languages. On laws the model had never seen, the right passage is among the top ten results for 93.6% of questions, up from 80.2%; on tax documents, 94.5% against 82.6%. On Belebele, the one test we have that was written by people, the gain is smaller and not confirmed (92.2% to 95.2%, with touching intervals), and Russian and English do not change. On law and tax questions written in the same style as its training questions, it is tied with Giga-Embeddings-3B, a model five times its size; on Belebele the 3B and 8B models score slightly higher, and only after we fine-tuned them on our data are they clearly ahead. We also report what failed: training in bfloat16 broke the model, and replacing Tajik letters with similar Russian ones did not help. Our law, tax and Wikipedia test questions come from the same model that wrote the training questions, so we treat those scores as an upper limit until questions written by Tajik speakers are available.

en
arXiv AI Papers

TVP: TV is Almost All You Need for Speculative Decoding

Speculative decoding accelerates LLM inference with a drafter whose proposed tokens are verified by the target model in parallel. Under lossless speculative sampling, a draft token is accepted with probability one minus the total variation (TV) distance between the target and draft distributions, yet drafters are often trained with cross-entropy (CE) or KL. We first show that under matched and well-tuned training, TV matches or surpasses several published draft-training objectives when fine-tuning, and a fixed TV and KL mixture does the same when training from scratch. We identify two remaining problems, TV's vanishing gradients on undercovered tokens and the prefix products that starve later positions in acceptance-length objectives. To address both, we derive TVP, a loss family based on expected acceptance length, whose power parameter interpolates between coverage and acceptance oriented objectives. Annealing the power from zero to one gives the highest or tied-highest acceptance length across three drafter architectures and three target models, including longer drafts and block verification. TVP improves acceptance length over CE by 9.8% when fine-tuning, and over the public DFlash2 training recipes by 6.2 to 12.9%.

en
arXiv AI Papers

Conic Certification of Transformer Attention

Attention is the computational core of large language models, yet the verifiers that bound how far an attention head's output can move under input perturbations give no guarantee on how loose their bounds are. A weak certificate therefore cannot be told apart from a fragile model. We develop a conic programming theory of attention certification that supplies this guarantee. We prove that the difficulty lies not in the softmax but in the bilinear interaction of a varying query with the keys. This interaction makes certification NP-hard over boxes and inapproximable once the attention logits can move by a constant amount. A fixed query or listed token substitutions, by contrast, make the exact worst case efficiently computable. An exponential cone moment relaxation then certifies a head with an error bounded quadratically in the perturbation radius by computable quantities, an order that cannot be improved. We also show that relaxations that treat a network one operation at a time, linear bound propagation among them, pay on second-order terms a sum of per-operation maxima instead of one coupled maximum. Computational results on GPT-2 indicate that the theory carries over to a trained model. Against exact worst cases with the query fixed, the certificate attains the predicted second-order accuracy. It is tighter than linear bound propagation on every head, and the looseness of bound propagation tracks the predicted per-operation excess.

en
arXiv AI Papers

WAR-CRAFT: Refining Web API Recommendations through Requirement Auditing

Web API recommenders return a set of APIs for a mashup request but do not check it against provider documentation, the main public evidence of an API's capabilities. The set can therefore contain unsupported or redundant APIs and miss requested capabilities. Filling every apparent gap does not fix this, because a gap may already be served, may not need its own API, or may have no documented provider. We study post-recommendation refinement, which revises a frozen recommender's output using only the request, the catalog, and provider documentation. We propose WAR-CRAFT, which treats refinement as three decisions. An LLM removes unsupported or subsumed APIs by judging candidates jointly, then audits the remaining set against a requirement checklist derived from the request. For each uncovered requirement, a fine-tuned catalog-constrained Resolver proposes a catalog API or abstains, and an independent documentation check admits only supported proposals. On ProgrammableWeb with 7 heterogeneous upstream recommenders, WAR-CRAFT raises mean SetF1 from 0.558 to 0.644, above the strongest of 5 refinement baselines (62.5%), and roughly halves functional redundancy (0.127 to 0.065) without changing mean functional coverage. Gains range from +2.2 points on generative recommenders to +18.2 on fixed-cutoff rankers. At matched set sizes, WAR-CRAFT still outperforms truncating the upstream output, so the gain is not explained by set-size calibration alone. Our code is available at https://github.com/ZsXu-enrico/WAR-CRAFT.

en
arXiv AI Papers

Can agents reliably design therapeutic antibodies?

Recent work has demonstrated agents driving protein design workflows, either through orchestration of specialized tools or direct manipulation of sequences. Yet the fidelity with which agents follow protein design instructions and the factors contributing to their effectiveness remain understudied. We examine agentic antibody design under two settings: when explicit molecular-editing instructions are provided and when agents receive only previously-scored sequences and are prompted to propose improvements. We call the first task-guided and the second black-box design. For the task-guided setting, we curate ten expert-validated affinity-improvement tasks and show that LLM judges closely reproduce programmatic evaluations of instruction following. We then study two augmentations: access to antibody-relevant tools and to previously scored design history. Tools improve instruction following (IF) and sequence space exploration while reducing optimization performance, whereas design history reduces IF but improves optimization performance at the expense of exploration. Finally, we combine taskguided and black-box design within an in silico optimization loop and show that our agentic optimizer performs comparably to an established Bayesian optimization baseline.

en
arXiv AI Papers

Distinguishing among reasoning levels of GPT 5.5 Pro via a grammar lens benchmark

Does the level of test-time processing effort affect a language model's computational capability? A formal grammar-lens benchmark method, derived from the theory of computation, generates novel data at specified levels, each successively more computationally powerful than the levels below. Using this data, we tested GPT-5.5, and GPT-5.5-Pro-Medium, -High, and -XHigh, showing incrementally increased acceptance of test data: 16, 40, 43, and 52\%, respectively, illustrating the benchmark's ability to discern fine-grained differences among models' added test-time reasoning levels. Additionally, as test-time compute effort increases, model responses diverge from human responses to the same data. This increasing disagreement may turn out to be broadly characteristic of more capable language models, introducing the potential of a general capability-versus-alignment conflict.

en
arXiv AI Papers

From Weeks to Hours: Fast and Principled SFT Data Generation for Reasoning LLMs

Reasoning-oriented post-training enables large language models (LLMs) to solve complex tasks via multi-step inference. Supervised fine-tuning (SFT) on high-quality reasoning traces is a particularly efficient approach, but its effectiveness depends critically on careful data curation. Existing pipelines rely on a costly generate-then-filter paradigm: large pools of long reasoning traces are produced by multiple teacher models and subsequently filtered for difficulty and diversity using additional LLMs, often requiring weeks of computation. In this work, we propose SFTProbe, a fundamentally different approach that bypasses this pipeline by directly predicting the quality of reasoning data. We show that a brief LoRA-based adaptation, combined with evaluating loss on only 5% of each reasoning trace, suffices to estimate both difficulty and diversity. SFTProbe is grounded in a geometric view of post-training, where pretrained LLMs lie in a low-rank, anisotropic loss basin. By probing this basin via low-rank perturbations, we construct compact loss signatures whose clustering captures diversity, while loss at curvature inflection points provides a robust proxy for difficulty. We also provide a simple criterion for teacher selection. Empirically, our approach reduces data curation time from weeks to hours while maintaining or improving performance. Remarkably, on mathematical reasoning, 5k generated examples match the highly filtered 89k-example in OpenThoughts-Math dataset, and 7.5k generated examples outperforms it by 2.5% in accuracy. We further introduce a high-quality physics reasoning dataset, and use it to train a model that outperforms an IPhO 2025 silver-medal-winning model. The code and the dataset are released at https://bigml-cs-ucla.github.io/SFTProbe-page/.

en
arXiv AI Papers

BitLLM: Efficient Low-Bit, Conversion-Hiding LLM Decoding on Mobile GPUs

The rapid scaling of large language models (LLMs) has unlocked remarkable capabilities across various domains. However, deploying LLMs on mobile devices remains challenging because of their high computational and memory requirements. Low-bit quantization (e.g., 4-bit weights, 8-bit activations) can reduce model size while preserving accuracy, but its practical efficacy on mobile GPUs has been limited so far. In this paper, we identify that the performance gap stems primarily from frequent data type conversions between quantized and floating-point values, and dynamic and inefficient memory access patterns caused by the growing Key-Value (KV) cache during autoregressive decoding. We introduce BitLLM, a mobile inference engine that addresses type conversion overhead and suboptimal memory accesses in quantized LLMs. Our approach features an adaptive layout and packing strategy for the KV cache that ensures efficient memory access throughout LLM decoding. This is coupled with an integer-native execution pipeline that defers and overlaps type conversion outside the compute hot path, keeping the critical MatMul stages in the integer domain. Implemented on off-the-shelf mobile GPUs, BitLLM outperforms leading engines, i.e., LLAMA.CPP, MNN-LLM, LiteRT-LM, and MLC-LLM, achieving average speedups of 1.78, 1.73, 1.58, and 1.47, respectively, across various models and quantization levels. Under long-context generation, BitLLM reduces average energy consumption (per token) by up to 88.9%.

en
arXiv AI Papers

Low-Margin Branch Points in Deterministic LLM Deliberatio

Deterministic multi-agent LLM deliberation can exactly re- play a fixed transcript, yet exact replay alone does not test local resistance to controlled internal interventions. Because each response becomes context for later agents, a changed greedy token can redirect a transcript-mediated execution. We present a white-box audit of this local failure mode in a five-agent committee under greedy decoding. We prospectively select low- and high-margin token positions, require exact no-op re- play, inject norm-relative residual displacements, and measure first transcript divergence within ten generated tokens. Low- margin positions are more branch-sensitive than high-margin controls, while matched margin-reducing gradient directions branch substantially more often than margin-increasing con- trols. We then continue selected realized branches through the committee to distinguish textual persistence from preference and final-decision changes. The contribution is a controlled audit for local branch susceptibility in this specified protocol, not an estimate of deployment frequency, decision correct- ness, or a phenomenon unique to multi-agent systems.

en
arXiv AI Papers

Skill++: Towards Lifelong Agents via Routing-Aware Skill Refinement

Large language model (LLM) agents extend their capabilities with skills, each with a name, a description, and a body. Under progressive disclosure, an LLM router selects skills by name and description alone and loads only the selected bodies. Skill libraries grow continually as users and agents add new skills. Because the router relies on this metadata, a similar new skill can displace an earlier one, breaking previously correct routing even though the earlier skill and the router are unchanged. This routing interference is aggravated for public skills, whose descriptions may not match their bodies. We formulate the problem as lifelong skill routing, a class-incremental continual learning problem. A solution must keep earlier skills selectable without suppressing new ones. Existing approaches fall short, since continually retraining the router risks catastrophic forgetting and rewriting methods either ignore similar skills or require per-skill feedback. We therefore propose Skill++, which rewrites only the name and description of each incoming skill, without router training or labeled data. The new description states what the body supports, how the skill differs from its closest predecessor, and what related work it excludes. As a library grows to 731 skills over ten sessions, Skill++ selects all required skills for 76.3% of queries, versus 36.4% under the original metadata. Across three benchmarks and five LLM routers, it outperforms all compared approaches, and its gains carry over to the Codex CLI agent harness.

en
arXiv AI Papers

Agentic Context Management for Multi-Turn Human-AI Conversations

LLMs degrade in multi-turn conversation partly because of context pollution: the model's own earlier outputs stay in context and anchor later turns, so an early misreading of the user's intent can persist after the user corrects it. Assistant omission (AO), dropping every assistant message, removes this pollution but fails when interaction is referential: later turns refer back to drafts, decisions, and tool-call results that exist only in assistant turns. We propose Agentic Conversation Context Curation (AC3), a training-free method in which a separate analyzer (1) consolidates the task specification from the user's messages alone, (2) audits the assistant's prior outputs against it, and (3) edits the context, either appending the analysis or discarding the assistant side and rebuilding the context from the specification. A second finding: pollution is contagious. An analyzer that sees the assistant's reasoning while extracting the specification can do worse than no intervention, and instructing it to ignore them did not close the gap, so we exclude assistant turns structurally. On Lost in Conversation (replaying difficulty-selected instances), where AO is an oracle by construction, AC3 closes 55-84% of the gap between full context and AO on three of four tasks and exceeds AO on the fourth, and an LLM that faithfully summarizes the history does not reproduce the gain even with more compute. On real WildChat conversations, a GPT-5-mini judge prefers AC3's next turn over both full context and AO in every configuration of our main table, though this preference depends on the judge; on CollabLLM results are mixed. In stateful tool use (tau2-bench), AO collapses to 0% on every model; AC3 does not collapse, though it falls slightly behind full context. Detector studies show that Reset removes injected pollution but is not span-selective: the gain comes from rebuilding the specification.

en
arXiv AI Papers

AStockArena: Beyond Returns, Can We Entrust Real Capital to the Decisions of Financial Agents?

Large language models (LLMs) are increasingly used in dynamic financial trading, yet real-time trading benchmarks mainly evaluate agents by cumulative returns and risk-adjusted performance. Such outcome metrics cannot show whether trades are faithfully reported, supported by available evidence, and consistent with account states. We introduce AStockArena, a real-time decision-auditing benchmark for continuous trading in China's A-share market. Under shared market conditions and execution constraints, we evaluate 13 open- and closed-source models over 3,198 sequential decisions. AStockArena reconstructsthe actions actually posted to accounts by aligning reports, trade requests, execution responses, and account states, and audits them with eight checks spanning execution traceability (O), evidence sufficiency (RK), and financial meta-awareness (B), combining record verification with consensus expert annotation. Of 1,070 trading points, 72.99% exhibit at least one audit gap; model-level rates range from52.2% to 91.2%, with evidence gaps most prevalent. Historical trade-filtering replay further shows thatportfolio outcomes are sensitive to these flagged trades. Decision-process auditing thus offers a verifiable complement to return-based evaluation of financial agents.

en
arXiv AI Papers

Do Social Influence Pressures Move LLM Agents? Evidence from Delegated Audit Judgment

Large language model (LLM) agents are rapidly being integrated into complex professional workflows such as auditing and entrusted with judgments under human review. Human auditors are known to yield to social influence pressures from superiors, peers and clients, and this paper asks whether LLM agents yield to the same pressures and, if so, how they respond. In an experiment on twelve model settings, an LLM agent acting as a senior associate in audit analytical review forms an estimate and is then challenged without new evidence by colleagues inside the firm, by a reviewer from another firm and by the client's CFO, with and without a career inducement. Experimental findings show that LLM agents yield to pressure from the engagement partner, from peers and from a reviewer from another firm, but in most settings not to pressure from the client's CFO. Under a career inducement the LLM agents revise their estimates less, not more, and they name the inducement as a threat to independence. Furthermore, some models revise only toward the more conservative estimate, while others revise in both directions. LLM agents thus do not respond to social influence pressures as human auditors do, and LLM-based decision support systems should be designed with the differences in mind.

en