
ALT Large language models demonstrate strong medical question-answering capabilities, yet reasoning performance remains uneven across languages. Process reward models (PRMs) provide step-level feedback for improving reasoning, but a fixed PRM may become increasingly misaligned with the trajectory distribution of an evolving policy during reinforcement learning, thereby reducing the reliability of its supervision. We propose a policyβreward alternating training framework that jointly adapts a medical PRM and optimizes a multilingual reasoning policy. Following supervised initialization, the current policy generates multilingual trajectories to update the PRM, which then provides process-level rewards for the next policy update. Medical step supervision maintains the PRM's medical foundation, while process, outcome, and language-related rewards guide policy learning. We develop multilingual medical PRMs (PAIR-PRM) and reasoning models (PAIR-MED) at 8B and 9B scales across 12 languages. On th

ALT KV cache compression is critical for large reasoning models, where most of the cost comes from long generated reasoning traces rather than the input prompt. These traces often contain redundant and low-utility tokens due to looping, self-verification, and corrections. However, such tokens may still receive high attention scores, making it difficult to remove them without harming the reasoning process. We propose RoCoKV, a KV cache compression method designed for large reasoning models. RoCoKV introduces two key components. First, uncertainty-guided adaptive pruning adjusts the compression strength based on the model's confidence, allocating more memory during uncertain phases and applying stronger pruning when the model is confident. Second, redundancy-aware token selection reduces the impact of tokens that are highly similar to recent reasoning content, encouraging the model to retain more informative and diverse tokens. Beyond the algorithm, we co-design RoCoKV with the serving stack

ALT Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow. Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with SkillPlumber, a framework in which an agent evolves its own skills where coverage shows they fall short. A persistent root cause analysis agent turns each label into a revision question and checks each revision against the next run, while a maintainer distills investigations across tasks into a shared strategy library. Across three benchmarks and three execu

ALT Long-horizon agents rely on stored records to recover a person's state as facts change. Reliable answers depend on how facts are recorded, retrieved, and presented to the reader. We treat a long-form persona dossier as a memory representation of a known latent state, an invariant identity plus a timeline of facts, routines, and decision rules tied to life phases. We present ChronoDossier, a benchmark of 311 synthetic dossiers of 1k-10k words written by seven generators from dated timelines. Intrinsic metrics evaluate the records against their intended states; extrinsic metrics test recovery by five readers on a seeded 32-persona GPT-5.1 panel balanced across lengths (520 rated turns per condition). The linked states, records, evidence packets, and answers support tests of construction and evidence delivery. In a contextual-stress control, tag-scoped writer inputs remove detected out-of-scope anchor copying. Frozen questions authored on staged dossiers transfer less reliably to one-pass

ALT Argumentation frameworks (AFs) provide explicit, formal mechanisms for representing and resolving conflicts and thus have been incorporated into large language model (LLM) decision-making. The benefits vary and depend on how arguments and their relations are constructed, evaluated, and incorporated into the decision process. We abstract a workflow consisting of four stages based on existing argumentation-augmented methods and investigate their design choices over tasks of debate-based decision-making and retrieval-based fact verification. Our evaluation with recent LLMs shows that existing methods do not consistently improve performance across the considered tasks. Motivated by this observation, we develop two task-specific methods. The first is called Qnt-Arg, which organizes arguments around the original speakers in debate materials and aggregates model-estimated argument strengths and candidate-answer priors under gradual semantics. The second is Qlt-Arg that uses credibility estima

ALT Language models are increasingly used to support human judgement in complex reasoning tasks, yet their predictions often remain difficult to interpret and validate. Explainability techniques provide insight into model decision-making, helping expose errors, failure modes, and potential bias so users can decide whether to rely on model outputs. Explanations are simplified approximations of a language model's internal workings. The goal is to obtain explanations that reflect the underlying computation as closely as possible (faithfulness), are understandable to users (plausibility), and help them achieve their intended goals (utility). Evaluation protocols are essential for assessing these dimensions. This thesis studies highlight explanations, natural language explanations, and chain-of-thought reasoning. These forms offer complementary perspectives on model behaviour and exhibit distinct strengths and limitations across tasks. Producing explanations that satisfy multiple quality dimens

ALT Zero-shot large language models (LLMs) are increasingly used as text classifiers, but their labels are only useful when we know to what extent we can trust them. We compare black-box methods (verbalised confidence and sampling) with white-box methods that read token probabilities (P(True), and multiple logit-scoring methods using constrained generation). We evaluate eleven open-weight LLMs (1B-32B parameters) on five classification datasets with 3 to 151 labels, measuring classification performance (accuracy, F1-macro), discrimination (AUROC), and calibration (ECE and Brier score). The choice of confidence method rarely impacts accuracy significantly, but logit-based scoring methods lead to much better discrimination (AUROC 0.74-0.77) than black-box methods (AUROC 0.57-0.66). Replacing labels with single-token identifiers preserves discrimination but costs 3-6 accuracy points on average. We also introduce decision-only scoring, which only scores positions where labels diverge in the pr

ALT Post-hoc rewriting is used to reduce demographic disparities in LLM-generated text when the source model is inaccessible. A rewrite can look successful merely because it improved its own target or satisfied its selector. We study this as a transfer problem on 660 gender pairs and 1,980 age contrasts of demographic-conditioned climate messages. A rewriter proposes edits, a selection model chooses among constraint-satisfying candidates, and an independent evaluator uninvolved in either rates pairwise framing disparity. A fixed policy that always targets the message with the lower Persuasion Bias Index (PBI) transfers inconsistently across axes and outcome aggregations. An adaptive policy that proposes edits on both sides but accepts at most one significantly improves on that rule, applied to the same candidates, in seven of eight comparisons across two evaluator families, mostly by editing more pairs, not by guided direction choice, at higher preservation risk. A policy editing both mess

ALT LLM judges serve as rewards where no answer key exists, in the hope that their knowledge passes to the trained model. In 814 RL runs that teach 7-8B models to answer closed-book questions or abstain, we ask what passes. What a grader knows about single answers stays home. A wrong answer that a judge rewards is memorised for that one question. A model trained with the answer key, or with a far stronger judge, chooses which new questions to answer no better than the untrained model does by sampling 32 answers and abstaining where they disagree; the grader's knowledge adds about one corrected answer per hundred questions. What a grader believes about a kind of question travels. A judge that rejects eleven of the twelve correct training answers about capitals of former states teaches the model to refuse 98% of such capitals on new questions, a loss that accuracy and abstention rates do not reveal. Changing only those twelve verdicts by hand moves the refusal from 5% to 90%, even for a grad

ALT Long-horizon LLM agents often exceed their context windows, forcing them to discard information that may be needed later. Existing context management strategies rely on summarization, masking, or structured memory, requiring decisions about what to preserve or how to retrieve it before future needs are known. However, these methods can introduce lossy compression, potentially discarding details needed for future tasks. To address this limitation, we introduce GrepMem, a simple, lossless, and index-free memory mechanism built on two components: an append-only plain-text log of the agent's complete history and a regular-expression search tool for on-demand recall. Unlike structured memory systems, GrepMem requires no embeddings, retrieval index, memory extraction, or additional models. We train Qwen3.5-9B with reinforcement learning under five context management strategies and evaluate on multi-question research sessions, long-stream recall, and agent-trajectory question answering. GrepM

ALT Large language models (LLMs) write fluent financial analyses but miscompute ratios, recall stale prices, and route questions to the wrong tool with high confidence. We ask which component should do each job in a financial question-answering assistant, and study it with FinVeriQA, an open-source multi-agent research prototype over SEC filings and live market data: a compact typed-decision model (Laya) routes each question and screens out advice requests, Python computes every number, retrieval and tools fetch evidence, Mistral-7B writes a cited answer, a code validator checks that cited IDs exist and that numbers match computed notes, and a Model Context Protocol (MCP) server exposes the pipeline. In live experiments, Laya routed better than Mistral-7B (0.85 vs. 0.52; overall 0.86 vs. 0.68, McNemar p = 0.0003); a 12B Mistral decided no better, and DeepSeek-R1-8B decided better (0.79 vs. 0.68, p = 0.0192) but about 9.9 times more slowly. Mistral-7B computed 29% of 100 financial calculati

ALT Efforts to keep clinical language models safe have concentrated on what enters and leaves the model, but a manipulated model can return an answer that looks entirely ordinary. Here we show that the model's own internal state, rendered in plain language, exposes manipulations the output conceals. Using a natural-language autoencoder, we translate the activations of a multimodal LLM (Gemma-3-27B) into text and let an input-blind judge read this verbalized internal state (the thoughts channel) and the model's answer separately. Across two clinical attack objectives (omitting a pathology, covertly recommending a drug), two input modalities, and two delivery mechanisms, the thoughts channel caught the injection significantly more often than the answer in all eight injected conditions (p<0.05), while both channels held near-zero false-positive rates on clean inputs; the omission result carried over to Llama-3.3-70B. The thoughts channel also registered unsafe clinical requests as harmful whi

ALT Populations vary along cultural dimensions such as individualism, collectivism, and honor. But how can these dimensions be studied in historical societies for which conventional psychological data are unavailable? Here, we introduce a computational text-analytic framework for quantifying psychological constructs in 54,670 Ottoman Turkish court records from 21 Istanbul courts spanning 1513-1911 CE. We develop sentence-embedding models adapted to Ottoman legal language and extend a theory-driven, text-analytic pipeline to capture cultural variation along the dimensions of individualism, collectivism, and honor in this historical corpus. We evaluate our models through multiple methods: document retrieval, semantic textual similarity, questionnaire-item classification, and validation against expert annotations. Our adapted sentence transformer (OTA-BERTurk) achieved a Recall@1 of.843 on the test set for document retrieval, a Mean Reciprocal Rank (MRR) of.806 for semantic textual similarity

ALT We present CANDID (Candidate-Driven Intervention for Disciplined Execution), an execution-control framework designed to enhance the reliability of local computer-use agents (CUAs) in resource-constrained settings. While local CUAs feature robust baseline proficiency in single-step action prediction, they frequently fail at critical junctions due to rigid action loops or premature DONE decisions. Rather than imposing imprecise cognitive loads through textual reflection, CANDID repurposes parallel test-time scaling into a disciplined, execution-level intervention at sparse, failure-prone points. We introduce a loop-recovery mechanism to filter candidates away from repeated behaviors and a conservative, unanimous DONE gating policy that requires absolute consensus to terminate. Experiments on OSWorld demonstrate that CANDID's structural control significantly improves success rates, outperforming feedback-based alternatives with minimal execution overhead.

ALT Jailbreak attempts and harmful prompts remain a challenge for the safe deployment of large language models. We study whether contrastively learned representations can support safety detection without training an additional downstream classifier. We propose Jailbreak Detection with Prototype Inference (JDPI), which uses binary safety labels to construct harmful and benign prototypes and makes predictions from relative prototype distances. JDPI uses only binary safety labels for encoder adaptation, prototype construction, and validation. In our setup, it uses fewer labeled examples than the classifier baseline (Method A) by avoiding the separate split used to train the downstream classifier. Across four benchmarks covering different input formats, contrastive fine-tuning improves JDPI's detection accuracy by 9.1-18.0 percentage points. Using the same encoder, JDPI achieves slightly higher ROC-AUC on one benchmark and stays within 0.017 of the baseline across all four. With dataset-specif

ALT Large language models have been widely used to generate plausible research ideas, yet it remains unclear what makes automated ideation effective and how generated novel proposals arise from prior scientific work. We introduce the Creative AI Researcher (CAR), an agentic framework combining literature grounding, iterative development and explicit conceptual operators. CAR constructs a semantic knowledge graph from retrieved literature and equips an ideation agent with graph-retrieval tools and four operators: bisociation, analogy, exploration and elaboration. Each operator call is checked against structural acceptance conditions and recorded, producing a trace of intermediate transformations and declared literature sources. We evaluate CAR with two language-model backbones across 40 LiveIdeaBench research topics and two execution benchmarks. CAR achieves significantly higher methodological novelty (MN) than single-shot prompting and The AI Scientist. Its methodological novelty does not

ALT Language models acquire their capabilities from training data, and developers withhold data during training to keep models from learning capabilities that could be misused. How long this delays a capability, and how much data is needed to acquire it later, is unknown. Here we show that the general progress a model makes on the rest of its training data substitutes for the data that was withheld. We withheld the training signal for individual skills in models of chess, Othello, molecular strings, bracket languages and web text. The delay of a withheld skill in larger models was predicted from smaller ones. A chess model never explicitly trained to respond to check ranked the legal responses almost as well as a model that was. When models were then trained on the withheld data, they acquired the skill with a small fraction of the data needed from initialization, and this fraction decreased as the models progressed. A control rule of equal difficulty was learned much more slowly. A progra

ALT Large language models (LLMs) are increasingly used to infer psychological attributes, predict individual behaviour, and simulate human populations. However, despite generating plausible responses, they often misrepresent individuals and distort population statistics. Here we introduce Chiron, a foundation model to measure and predict human behaviour. We jointly fine-tune an LLM to represent psychological constructs, predict individual behaviour, and align population response distributions with human statistics. The training and evaluation data span 36 behavioural domains and 78 social science experiments, comprising 288,307 person records, 1,411,750 atomic behavioural decisions, and 99,953 psychometric assessments over 109 psychological constructs. In our evaluations, Chiron's psychological measurements agree more closely with human self-reports and exhibit fewer biases that undermine psychometric validity. It predicts individual behaviour more accurately than frontier LLMs (up to 28.4

ALT Hybrid attention models combine full and linear attention to achieve a better qualityβefficiency trade-off. Converting pretrained Transformers into such models requires identifying which layers should retain full attention under a prescribed budget. We identify two off-policy mismatches in FlashMorph's layer selection: (i) continuous branch interpolation and teacher-forced hidden states differ from discrete hybrid computation; and (ii) frozen model weights fail to account for parameter adaptation during subsequent recovery. We introduce FlashMorph2, an on-policy layer-selection method that (I) evaluates sampled discrete hybrid architectures end to end using their own hidden states, and (II) jointly optimizes layer-selection scores and model weights under the recovery objective to reflect post-adaptation performance. Experiments across Qwen3 model scales and linear-attention variants show that FlashMorph2 consistently outperforms existing layer-selection methods, improving average score

ALT What makes a small trained map from encoder states to the embedding space of a frozen language model sufficient for competitive speech recognition? The premise of the SLAM-ASR architecture is that a frozen language model possesses linguistic knowledge rich enough to perform the recognition task. This assumption, while reasonable, provides little insight into the mechanisms underlying these models. We therefore seek the concrete internal structures on which they rely. This work shows that SLAM-ASR models rely on induction heads of the frozen language model: the same heads that copy repeated token sequences in text are also important for mapping speech embeddings to transcript tokens. Moreover, we find pairs of heads that behave as a constrained form of differential attention, cancelling attention noise and sharpening attention onto the previously generated word. Finally, we show how induction heads could be used to improve transcription quality under noisy audio.
