Daily Papers

Recent transformer-based approaches have demonstrated excellent performance in 3D human pose estimation. However, they have a holistic view and by encoding global relationships between all the joints, they do not capture the local dependencies precisely. In this paper, we present a novel Attention-GCNFormer (AGFormer) block that divides the number of channels by using two parallel transformer and GCNFormer streams. Our proposed GCNFormer module exploits the local relationship between adjacent joints, outputting a new representation that is complementary to the transformer output. By fusing these two representation in an adaptive way, AGFormer exhibits the ability to better learn the underlying 3D structure. By stacking multiple AGFormer blocks, we propose MotionAGFormer in four different variants, which can be chosen based on the speed-accuracy trade-off. We evaluate our model on two popular benchmark datasets: Human3.6M and MPI-INF-3DHP. MotionAGFormer-B achieves state-of-the-art results, with P1 errors of 38.4mm and 16.2mm, respectively. Remarkably, it uses a quarter of the parameters and is three times more computationally efficient than the previous leading model on Human3.6M dataset. Code and models are available at https://github.com/TaatiTeam/MotionAGFormer.

3D human pose estimation is a classic and important research direction in the field of computer vision. In recent years, Transformer-based methods have made significant progress in lifting 2D to 3D human pose estimation. However, these methods primarily focus on modeling global temporal and spatial relationships, neglecting local skeletal relationships and the information interaction between different channels. Therefore, we have proposed a novel method,the Dual-stream Spatio-temporal GCN-Transformer Network (MixTGFormer). This method models the spatial and temporal relationships of human skeletons simultaneously through two parallel channels, achieving effective fusion of global and local features. The core of MixTGFormer is composed of stacked Mixformers. Specifically, the Mixformer includes the Mixformer Block and the Squeeze-and-Excitation Layer ( SE Layer). It first extracts and fuses various information of human skeletons through two parallel Mixformer Blocks with different modes. Then, it further supplements the fused information through the SE Layer. The Mixformer Block integrates Graph Convolutional Networks (GCN) into the Transformer, enhancing both local and global information utilization. Additionally, we further implement its temporal and spatial forms to extract both spatial and temporal relationships. We extensively evaluated our model on two benchmark datasets (Human3.6M and MPI-INF-3DHP). The experimental results showed that, compared to other methods, our MixTGFormer achieved state-of-the-art results, with P1 errors of 37.6mm and 15.7mm on these datasets, respectively.

Large language models can diagnose a student program from its code, problem statement, and reference solution. Evaluating this ability requires a clear definition of what counts as the correct diagnosis. We introduce PyMETA, a Python error dataset with 48,646 student submissions to 155 problems. Every submission has a single label for the first execution error identified by an Online Judge, or No Error when the program passes all tests. A targeted subset of 97 submissions also has expert labels collected through iterative repair and re-execution. The taxonomy has three levels; its most detailed level contains 14 labels, including No Error, Logic Error, named Python exceptions, and an Other Errors category. We evaluate two finetuned models and two groups of prompted LLMs: four earlier models and four recent models. When evaluated against the first execution error, the recent prompted models reach 87.5--93.8% macro F1, above the strongest finetuned baseline at 80.6%. This is the opposite of the comparison obtained with the earlier prompted models. On the 97-item expert subset, however, exact-set match is only 43.3--48.5%, although sample F1 is about 79--81%. Output format also matters. On the same 45 audited submissions whose expert label sets do not contain Logic Error, none of the four recent models returns that label under single-error prompting, but 46.7--57.8% of their multi-error outputs include it, usually after an explicit-error label. The results show that model rankings and claims about label bias depend on the meaning of the gold label, the number of labels a model may return, and the scoring rule.

Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric. Physics is the sharpest test of this shift, which binds symbols to reality in a fundamental way, serving as the cornerstone of most modern technologies. In this work, we manage to advance physics research by developing large language models with exceptional physics reasoning capabilities, especially excel at solving Olympiad-level physics problems. We introduce P1, a family of open-source physics reasoning models trained entirely through reinforcement learning (RL). Among them, P1-235B-A22B is the first open-source model with Gold-medal performance at the latest International Physics Olympiad (IPhO 2025), and wins 12 gold medals out of 13 international/regional physics competitions in 2024/2025. P1-30B-A3B also surpasses almost all other open-source models on IPhO 2025, getting a silver medal. Further equipped with an agentic framework PhysicsMinions, P1-235B-A22B+PhysicsMinions achieves overall No.1 on IPhO 2025, and obtains the highest average score over the 13 physics competitions. Besides physics, P1 models also present great performance on other reasoning tasks like math and coding, showing the great generalibility of P1 series.

We investigate how looped transformers encode human preference, training lightweight evaluator heads on frozen Ouro-2.6B loop-iteration states on Anthropic HH-RLHF. v2: an erratum is prepended; the original manuscript is unchanged. A post-publication audit found the three headline results inflated by two independent evaluation errors. The 95.2% pairwise evaluator accuracy is a canonical-ordering artifact: the data were correctly split, but the evaluator learned to prefer the first-presented argument; its strict antisymmetrized accuracy on the full 8,552-pair test set is 63.9%. The 84.5% pairwise probe and the below-chance 21.75% pointwise probe were source-item leaks (orientation rows and pair partners crossing the train/test split); corrected pair-disjoint values are 56.5% and 54.2% -- above chance, so the "inverted polarity" finding is withdrawn. The central finding survives at much smaller magnitude: preference is decoded more accurately relationally than pointwise (paired +2.3 points, 95% CI [+1.3, +3.3]), and the antisymmetrized evaluator still beats the linear probe, but no corrected readout rivals end-to-end reward models. The methodological findings stand (constant-output degeneracy, flip test, swap-protocol metric deflation), with one correction: antisymmetrized accuracy, not antisymmetry correlation, certifies relational discrimination. The two errors are mutually invisible -- a split audit cannot see an ordering prior, antisymmetrization cannot see a leak -- so both checks are required. Full audit in the follow-up work (Kirin, 2026, in preparation).

Many errors in student essays can be explained by influence from the native language (L1). L1 interference refers to errors influenced by a speaker's first language, such as using stadion instead of stadium, reflecting lexical transliteration from Russian. In this work, we address the task of detecting such errors in English essays written by Russian-speaking learners. We introduce RILEC, a large-scale dataset of over 18,000 sentences, combining expert-annotated data from REALEC with synthetic examples generated through rule-based and neural augmentation. We propose a framework for generating L1-motivated errors using generative language models optimized with PPO, prompt-based control, and rule-based patterns. Models fine-tuned on RILEC achieve strong performance, particularly on word-level interference types such as transliteration and tense semantics. We find that the proposed augmentation pipeline leads to a significant performance improvement, making it a potentially valuable tool for learners and teachers to more effectively identify and address such errors.

Autonomous agents increasingly perform bounded software tasks under an orchestrator that retries, resumes, and budgets them. The machinery such orchestrators reach for is the service mesh's: retry, timeout, and error-rate circuit breaking. We report a failure study of a production agentic software-delivery platform over 147 numbered incidents spanning 81 runs, each with a measured cost and, in most cases, a mutation proof reproducing the failure. All three assumptions those primitives rest on are violated in practice, and we quantify the consequences: a loop of fifty-four consecutive successful tool calls no error-rate breaker could see; a progress signal constant by construction, guaranteeing a false trip on the third repair round and driving one run from six of six components to three; twenty-one events accumulated across six invocations of one delegation, making a correct, idempotent component unwinnable; a misrouted failure that woke five components for a two-component fault, leaving three bystanders regressing working code; and twelve incidents in which the enforcement layer blocked correct work, the most expensive costing 107 agent turns and zero accepted writes. We find one cross-cutting cause and its dual. Identity adequacy: in five separate subsystems an identity that failed to discriminate produced a confident wrong answer, and two of them derived the corrective rule independently. Evidence adequacy: a reliability decision may be taken only on evidence capable of moving, attributable to what it measures, and deterministic under identical conditions. From the findings we derive seven reliability primitives whose enforcement unit is the delegation rather than the message, and specify the controlled evaluation the study motivates but does not constitute.

Background: The positive predictive value (PPV) of large language model (LLM)-based proofreading for radiology reports is limited due to the low error prevalence. Purpose: To assess whether a three-pass LLM framework enhances PPV and reduces operational costs compared with baseline approaches. Materials and Methods: A retrospective analysis was performed on 1,000 consecutive radiology reports (250 each: radiography, ultrasonography, CT, MRI) from the MIMIC-III database. Two external datasets (CheXpert and Open-i) were validation sets. Three LLM frameworks were tested: (1) single-prompt detector; (2) extractor plus detector; and (3) extractor, detector, and false-positive verifier. Precision was measured by PPV and absolute true positive rate (aTPR). Efficiency was calculated from model inference charges and reviewer remuneration. Statistical significance was tested using cluster bootstrap, exact McNemar tests, and Holm-Bonferroni correction. Results: Framework PPV increased from 0.063 (95% CI, 0.036-0.101, Framework 1) to 0.079 (0.049-0.118, Framework 2), and significantly to 0.159 (0.090-0.252, Framework 3; P<.001 vs. baselines). aTPR remained stable (0.012-0.014; P>=.84). Operational costs per 1,000 reports dropped to USD 5.58 (Framework 3) from USD 9.72 (Framework 1) and USD 6.85 (Framework 2), reflecting reductions of 42.6% and 18.5%, respectively. Human-reviewed reports decreased from 192 to 88. External validation supported Framework 3's superior PPV (CheXpert 0.133, Open-i 0.105) and stable aTPR (0.007). Conclusion: A three-pass LLM framework significantly enhanced PPV and reduced operational costs, maintaining detection performance, providing an effective strategy for AI-assisted radiology report quality assurance.

In many binary classification applications such as disease diagnosis and spam detection, practitioners often face great needs to control type I errors (i.e., the conditional probability of misclassifying a class 0 observation as class 1) so that it remains below a desired threshold. To address this need, the Neyman-Pearson (NP) classification paradigm is a natural choice; it minimizes type II error (i.e., the conditional probability of misclassifying a class 1 observation as class 0) while enforcing an upper bound, α, on the type I error. Although the NP paradigm has a century-long history in hypothesis testing, it has not been well recognized and implemented in classification schemes. Common practices that directly limit the empirical type I error to no more than α do not satisfy the type I error control objective because the resulting classifiers are still likely to have type I errors much larger than α. As a result, the NP paradigm has not been properly implemented for many classification scenarios in practice. In this work, we develop the first umbrella algorithm that implements the NP paradigm for all scoring-type classification methods, including popular methods such as logistic regression, support vector machines and random forests. Powered by this umbrella algorithm, we propose a novel graphical tool for NP classification methods: NP receiver operating characteristic (NP-ROC) bands, motivated by the popular receiver operating characteristic (ROC) curves. NP-ROC bands will help choose α in a data adaptive way and compare different NP classifiers. We demonstrate the use and properties of the NP umbrella algorithm and NP-ROC bands, available in the R package nproc, through simulation and real data case studies.

Lean verifies that a generated declaration is well typed, but cannot check whether it states what the user intended. We measure how far compilation overstates faithfulness in statement autoformalization without canonical Lean targets. Our criterion requires Lean compilation and agreement of two LLM judges from different model families. On an independent random audit, it agrees with the human majority on 91.5% of cases; it rarely accepts an unfaithful output but rejects some faithful ones. Across eight systems on 227 university-level statements, the gap between compiling and accepted outputs ranges from 1.3 to 29.5 percentage points: our tool-augmented agent compiles 87.2% of the statements, but only 57.7% are accepted. Human review of all the agent's rejections confirms that 23.8 of its 29.5 points are real translation errors. For the gaps of the seven one-shot systems, a two-stage review by Claude Opus 5.5 with Lean and GPT-6.1 Sol, validated on these human labels, finds that 78-100% are real translation errors for six of them. When GPT-5.2, one of the two judges, grades the translations of two models released after it, GPT-6 Astra and Claude Opus 5.5, it accepts 83% of the statements for each, against 21% for one-shot GPT-5.2. For these two models, the same review classifies only 4 of their 54 rejections (7%) as translation errors, so most of their measured gap is judge errors rather than translation errors. This is consistent with these systems erring less while the judge does not: the fewer errors a system makes, the smaller the share of its rejections that are translation errors. Finally, a 2^3 tool ablation of the agent shows that Lean feedback drives most of its compilation gain, and that nearly half of this gain fails the semantic criterion. Code and data are available at https://github.com/grenadeComing/beyond-compilation.

Ambient AI scribes draft clinical notes, and published audits find their dominant error is omission: information the encounter established that the note fails to record. The standard check is an LLM judge: a second model reads the note against the transcript and flags problems. We ask whether judges detect omissions. Public corpora cannot supply the answer key: their clinician reference notes and transcripts are materially discrepant. Our benchmark has 500 single-error note pairs from audited fact sheets, 298 with a named fact certainly absent and 202 added-or-altered controls. Across eight judge designs, paired discrimination (the flawed note below its clean twin, 0.5 a coin flip) reads 0.79-0.94 on added or altered content and 0.50-0.63 on omissions. On single notes, no design flags omissions reliably more often than perfect notes. Wording changes, voting and GEPA prompt optimisation move the operating point without creating usable detection. Restructuring the task recovers it: list the facts the transcript establishes, then check the note for each. Two methods reach it independently and trade off: a per-fact pipeline, and a GEPA-evolved prompt doing the same in one call. The pipeline's flags name the missing fact and its severity at 2.7% false alarms. The single call detects more (36.9% against 24.6%, p=0.002) at 6.2% false alarms and a tenth of the cost per note. A physician author validated 70 items and, where the two routes disagree, sided with the pipeline on 10 of 10 (p=0.002). A second clinician, not an author, graded the severity rubric blind and agrees to within a grade. On real vendor notes from a companion census no benchmark threshold transfers, but the re-calibrated single call detects more than the best of the eight at half its false-alarm rate. Omissions whose fact is restated elsewhere defeat both routes. We release the benchmark, prompts and judgements.

We introduce a new type of programming challenge called programming puzzles, as an objective and comprehensive evaluation of program synthesis, and release an open-source dataset of Python Programming Puzzles (P3). Each puzzle is defined by a short Python program f, and the goal is to find an input which makes f return True. The puzzles are objective in that each one is specified entirely by the source code of its verifier f, so evaluating f is all that is needed to test a candidate solution. They do not require an answer key or input/output examples, nor do they depend on natural language understanding. The dataset is comprehensive in that it spans problems of a range of difficulties and domains, ranging from trivial string manipulation problems, to classic programming puzzles (e.g., Tower of Hanoi), to interview/competitive-programming problems (e.g., dynamic programming), to longstanding open problems in algorithms and mathematics (e.g., factoring). We develop baseline enumerative program synthesis, GPT-3 and Codex solvers that are capable of solving puzzles -- even without access to any reference solutions -- by learning from their own past solutions. Codex performs best, solving up to 18% of 397 test problems with a single try and 80% of the problems with 1,000 tries per problem. In a small user study, we find a positive correlation between puzzle-solving performance and coding experience, and between the puzzle difficulty for humans and AI solvers. Therefore, further improvements on P3 could have a significant impact on many program synthesis areas.

We describe our system for LitTraceQA (GroundLM @ EMNLP 2026): given a research question, retrieve the relevant papers from a pool of 27,487, cite the page and the table or figure where the answer lives, and answer in a requested format. Our main finding is that evidence grounding is limited by retrieval, not by reading. The page selector put the annotator's page, which we call the gold page, in front of the model that locates evidence only about half the time (52.6% gold-page recall), while that model, given the page, cited the right one in 45 of the 48 locators it emitted (94%). When the page was missing it rarely said so: of 45 such cases it returned nothing 14 times, a wrong page 24 times, and a correct page 7 times, so the pipeline failed quietly almost twice as often as it failed visibly. Since the failure was that the right page was never shown, the fix is to stop choosing: each retrieved paper fits in the model's context, so we show it whole. Page ranking survives only as a fallback inside papers too long to fit, which no test-split paper was, and gold-page recall reaches 100% on the papers we can parse. Separately, questions that identify their target by position rather than content, such as "the first author of the 24th reference", are served by parsing rather than retrieval: we resolve the bibliography into an addressable list, which also supplies identifiers the evidence metric scores. The final system scores 0.762 paper F_1, 0.441 evidence F_1 and 0.920 multiple-choice accuracy on the held-out test split. Because the pipeline depends on a closed model without seed control, we release a harness that verifies the paper's central claims against committed artifacts.

Reproducing a machine learning paper involves most research steps, from installing software and debugging to running experiments, work that AI agents increasingly do. We introduce RECLAIM, a benchmark of 100 NeurIPS 2025 papers that can be rebuilt yearly from new conferences. For each paper we fix in advance the result to reproduce, what counts as a successful reproduction, and a GPU-hour budget. An agent must reproduce that result using the paper and whatever its authors released. What the authors released decides the difficulty tier. Run-tier releases include code, data, and weights; Retrain-tier releases lack weights, so the agent trains the model; Reimplement-tier releases lack code, so the agent writes it. A separate language model grades runs from logs and outputs rather than agents' reports. We run four agents once per paper; the best agent in each tier reproduces only 41% of Run-tier papers, 27% at Retrain, and 15% at Reimplement, where every agent does worst. Failed attempts use on average 29% of their budget, so most stop with budget left. The most common agent error is writing the method without checking any part against the paper's numbers, in 63 of 400 runs.

Majority voting over multiple LLM attempts improves mathematical reasoning, but correlated errors limit the effective sample size. A natural fix is to assign different reasoning strategies to different voters. The approach, Diverse Prompt Mixer, is tested on the AIMO 3 competition: 3 models, 23+ experiments, 50 IMO-level problems, one H100 80 GB, 5-hour limit. Every prompt-level intervention fails. High-temperature sampling already decorrelates errors; weaker strategies reduce accuracy more than they reduce correlation. Across an 8-point capability gap at equal N=8 and every optimization tested, model capability dominates. The gap between the best majority-vote score (42/50) and pass@20 (~45.5) is selection loss, not prompt loss. A verifier-based selector could close it. Prompt engineering cannot.

Many datasets contain personally identifiable information, or PII, which poses privacy risks to individuals. PII masking is commonly used to redact personal information such as names, addresses, and phone numbers from text data. Most modern PII masking pipelines involve machine learning algorithms. However, these systems may vary in performance, such that individuals from particular demographic groups bear a higher risk for having their personal information exposed. In this paper, we evaluate the performance of three off-the-shelf PII masking systems on name detection and redaction. We generate data using names and templates from the customer service domain. We find that an open-source RoBERTa-based system shows fewer disparities than the commercial models we test. However, all systems demonstrate significant differences in error rate based on demographics. In particular, the highest error rates occurred for names associated with Black and Asian/Pacific Islander individuals.

A large share of day-to-day work in professional domains happens inside PDF files: benefits packets, leases, datasheets, clinical guidelines, construction plans. Benchmarks for document AI have generally measured the required capabilities in isolation: OCR, layout analysis, chart reasoning, table QA, document VQA. A high score on any one of them does not necessarily reveal whether a model can answer a realistic question that someone in the field would actually ask about a specific PDF. GDP_pdf is a benchmark built to measure this directly. It consists of question-document pairs authored by working professionals in ten fields, and a candidate question was kept only when at least two frontier multimodal models failed it in a way that mattered: a wrong answer, missed decisive evidence, or a fabricated claim, rather than a superficial difference such as style. Each item comes with a rubric of atomic criteria, so we can report a graded rubric score as well as a strict task-level pass rate, and each item is tagged against a taxonomy of eleven capabilities in three tiers, spanning text extraction and grounding, table and chart comprehension, cross-referencing, spatial reasoning, and abstention on unsupported queries. We report results for seventeen frontier models on the 100-item benchmark: the best model passes only 30.7% of the items and the worst passes 2%. Most errors trace back to a small set of recurring loss patterns: misaligned tables, misread charts, skipped footnotes and exclusions, miscounted floor-plan symbols, scan noise, and amendments that supersede earlier text.

Chain-of-Thought (CoT) prompting methods have enabled large language models (LLMs) to generate reasoning paths and solve math word problems (MWPs). However, they are sensitive to mistakes in the paths, as any mistake can result in an incorrect answer. We propose a novel method named Progressive Rectification Prompting (PRP) to improve average accuracy on eight MWP datasets from 77.3 to 90.5. Given an initial answer from CoT, PRP iterates a verify-then-rectify process to progressively identify incorrect answers and rectify the reasoning paths. With the most likely correct answer, the LLM predicts a masked numerical value in the question; if the prediction does not match the masked value, the answer is likely incorrect. Then the LLM is prompted to re-generate the reasoning path hinted with a set of incorrect answers to prevent itself from repeating previous mistakes. PRP achieves the best performance compared against the CoT methods. Our implementation is made publicly available at https://wzy6642.github.io/prp.github.io/.

We consider the task of mapping pseudocode to long programs that are functionally correct. Given test cases as a mechanism to validate programs, we search over the space of possible translations of the pseudocode to find a program that passes the validation. However, without proper credit assignment to localize the sources of program failures, it is difficult to guide search toward more promising programs. We propose to perform credit assignment based on signals from compilation errors, which constitute 88.7% of program failures. Concretely, we treat the translation of each pseudocode line as a discrete portion of the program, and whenever a synthesized program fails to compile, an error localization method tries to identify the portion of the program responsible for the failure. We then focus search over alternative translations of the pseudocode for those portions. For evaluation, we collected the SPoC dataset (Search-based Pseudocode to Code) containing 18,356 programs with human-authored pseudocode and test cases. Under a budget of 100 program compilations, performing search improves the synthesis success rate over using the top-one translation of the pseudocode from 25.6% to 44.7%.

LLMs have long demonstrated remarkable effectiveness in automatic program repair (APR), with OpenAI's ChatGPT being one of the most widely used models in this domain. Through continuous iterations and upgrades of GPT-family models, their performance in fixing bugs has already reached state-of-the-art levels. However, there are few works comparing the effectiveness and variations of different versions of GPT-family models on APR. In this work, inspired by the recent public release of the GPT-o1 models, we conduct the first study to compare the effectiveness of different versions of the GPT-family models in APR. We evaluate the performance of the latest version of the GPT-family models (i.e., O1-preview and O1-mini), GPT-4o, and the historical version of ChatGPT on APR. We conduct an empirical study of the four GPT-family models against other LLMs and APR techniques on the QuixBugs benchmark from multiple evaluation perspectives, including repair success rate, repair cost, response length, and behavior patterns. The results demonstrate that O1's repair capability exceeds that of prior GPT-family models, successfully fixing all 40 bugs in the benchmark. Our work can serve as a foundation for further in-depth exploration of the applications of GPT-family models in APR.

Agentic on-policy distillation (OPD) trains a student agent on its own trajectories while a teacher scores every token. A natural explanation of where it goes wrong is compounding error: the student's mistakes carry its trajectories beyond the region where the teacher is reliable, so recent methods keep training inside that region, through episode-length curricula, teacher-guided turns, prefix replay, or teacher intervention where the two disagree. We show this explanation is incomplete. From the states where the student departs most from the teacher, the teacher's own takeovers still finish the task, yet the student trained on its token scores keeps running out of turns. We attribute this to the myopia of next-token scoring. The teacher rates each token without seeing where the episode ends, so its scores let through the safe, delaying moves that outcome-based RL would punish with zero reward, a penalty agentic OPD lacks. We call this difference between what the teacher does and what it scores the rollout-scorer gap. We propose Rollout-to-Scorer Distillation (R2S), which repairs the scorer before distillation by mining high-disagreement states from the untrained student, running teacher takeovers from them, keeping the states where enough takeovers pass a reward-free check such as consensus or an LLM judge, and fine-tuning the teacher to give its takeover-informed scores on every assistant token from the divergence turn onward, without the takeover in context. The repaired teacher is frozen as the sole scorer for standard OPD, requiring no task reward during repair or training. Across three environments, two teachers and four students, R2S outperforms three state-of-the-art agentic-OPD methods in all twelve teacher-student-environment cells, by 12.8 success points on average over the best of them in each cell. Project page: https://tree-shu-zhao.github.io/r2s/.

An agent that inherits six one-line memories may pull at most one archived source record before acting; a directive written into the store can steer that choice: a pointer to the record, a criterion that identifies it, or both. Across twelve registered studies on one instrument lineage (14,760 attempts) we measured where the request goes under each form. On six direct-provider models a length-matched criterion exceeded a bare id by +35.0 points [+31.2, +38.8] (Study D); the contrast failed its registered superiority rule on a nine-model OpenRouter-served panel (Study E). Appending the id cancelled the criterion on three Claude models (Opus 5: 40/40 to 0/40; Study F-x); six byte-matched edits gave each exact string its own effect (Study G), and a re-run at eighty runs per cell left fifteen of thirty replication contrasts within the margin, fifteen unresolved and none beyond (Study G'). A ratification line (+96.0 points on Opus 5) and a budget of two credits restored the target on all three (Study J); across five criterion strings the suffix's cancellation held for four of the five wordings on Opus 5 and all five wordings on Fable 5.1 (Study H2); in a second store every model followed the criterion (Study H1). Continued into a decision, the criterion moved the choice toward the current record (+100.0 points, Opus 5) and away from it on Fable 5.1 (Study I). A one-character plan pointer's effect (+78.0 points; Study B, after a correction of its first repository report) returned the same verdict under a prospectively registered re-run (+81.7 points; Study B'). All results are descriptive effects of exact edits on fixed panels with registered intervals and no mechanism claim.

Benchmark contamination is diagnosed with n-gram overlap, likelihood-based membership inference, or canary strings, and each needs something usually unavailable: the training corpus, a well-chosen test statistic, or foresight at release. A recent alternative reads it off a linear probe on internal activations. We show the natural way to do this does not work, specify one that survives measurement, then find that the correction making it work carries more variance than the null it is tested against. The protocol reports a zero-sum contrast on the depth profile of probe accuracy, recentred on a level-matched placebo baseline, tested against a label-permutation null, with the reference set twice the size of the suspect set. Each choice replaces a simpler alternative we rejected on measurement. Reporting the level of excess separability rather than its shape makes the false positive rate track the size of the analyst's own control set, 0.03 to 0.99 under a true null. Contrasting against a flat depth profile rejects a true null 0.72 of the time when surface decodability rises with depth, and loses all power when it falls. On real transformers the protocol fails a test the simulations did not pose. The recentring subtracts an estimate, and the permutation null holds it fixed. Re-estimated across split seeds on four audits of contaminated checkpoints, its standard deviation is 1.30 to 1.56 times the null's own in every arm: what is subtracted to remove a bias is more variable than what it corrects. The one nominally significant result, p = 0.0075, becomes 0.0745 once that variance is propagated, and no verdict is issued. The simulations missed this because their surface key is the covariate driving item variation; on real text it is a proxy, and degrading key quality in simulation reproduces it. We add a companion measurement and a widened null. No arm shows contamination.

This paper presents a case study of coding tasks by the latest reasoning models of OpenAI, i.e. o1-preview and o1-mini, in comparison with other frontier models. The o1 models deliver SOTA results for WebApp1K, a single-task benchmark. To this end, we introduce WebApp1K-Duo, a harder benchmark doubling number of tasks and test cases. The new benchmark causes the o1 model performances to decline significantly, falling behind Claude 3.5. Moreover, they consistently fail when confronted with atypical yet correct test cases, a trap non-reasoning models occasionally avoid. We hypothesize that the performance variability is due to instruction comprehension. Specifically, the reasoning mechanism boosts performance when all expectations are captured, meanwhile exacerbates errors when key expectations are missed, potentially impacted by input lengths. As such, we argue that the coding success of reasoning models hinges on the top-notch base model and SFT to ensure meticulous adherence to instructions.

Many NP-complete (NPC) problems have a polynomial-time (P) twin that differs in a single structural detail, such as 3- versus 2-colouring a graph. A model that reasons rather than recalls should notice which side of this boundary a problem falls on, yet existing benchmarks either fix the wording of each task or ignore its complexity class, so they cannot separate the two. We introduce P2NPCBench, a benchmark of 12 problem families in which every NPC problem is paired with a P twin posed over the same surface. To remove the textbook cues that enable recall, we build it with AgentGAN, an adversarial multi-agent framework in which a generator rewrites each statement until a panel of frozen code models fails, while an executable validity gate keeps every rewrite answer- and class-preserving. The result is 2,114 hardened statements with hand-verified reference human solvers. Evaluating 12 proprietary and open-source models, we find that every model scores lower on NPC statements than on their P twins (35.4% vs. 63.8% pass@1), and that 95.1% of model-family pairs fail more often on the NPC side. A tenfold time limit raises NPC accuracy only from 28.5% to 31.9%, so failures stem from wrong or crashing programs rather than slow ones. Hardening also narrows the class gap from 24.8 to 14.6 points, showing that part of the apparent gap reflects familiarity with textbook phrasing rather than complexity class.

Ambient AI scribes draft clinical notes under the reassurance that a clinician signs every note. We audited three commercial AI scribes on the same 142 consultations: 565 notes from recorded UK primary-care and US ambulatory encounters plus authored scenarios. Twelve discovery passes proposed 13,678 candidate errors; the 5,898 clearing an importance filter went to an adversarial panel of two models from different families, each told to refute what it could, and 618 survived. One note in three (31.3% [27.0, 35.6]) carries a verified failure, concentrated in allergy and medication information, invented patient identity, and history written up as examination on telephone consultations that can contain none. No product was given a patient record; setting aside the two classes a record would have prefilled, invented identity and dates, the rate is 24.8% [20.8, 29.0]. One failure mode did not fit our scheme, drawn from published scribe-error taxonomies: a treatment the clinician retracts, recorded as delivered care. Two clinicians adjudicated blind, disjoint samples: a physician author upheld 20 of 21 findings (95.2% [77.3, 99.2]) and an independent clinician, not an author, 12 of 12 ([75.8, 100]); both judged every sampled refusal genuine. A failure rate depends on the instrument as much as the scribes. With model, evidence and settings fixed, the review instruction alone moves the share of candidates verified from 9.3% to 79.0%, and the reviewing family moves it too: alone at that instruction the gentler flags 54.8% of notes against 27.8%. Between 28% and 97% of sampled notes carry a failure depending on the standard. Published audits disagree among themselves by a margin instrument differences alone can produce: omission is 54-86% of their errors against our 23.1%. We release all 618 findings with transcript-side evidence, every prompt and model version, and the re-runnable pipeline.

Recent research has generated hope that inference scaling could allow weaker language models to match or exceed the accuracy of stronger models, such as by repeatedly sampling solutions to a coding problem until it passes unit tests. The central thesis of this paper is that there is no free lunch for inference scaling: indefinite accuracy improvement through resampling can only be realized if the "verifier" (in this case, a set of unit tests) is perfect. When the verifier is imperfect, as it almost always is in domains such as reasoning or coding (for example, unit tests have imperfect coverage), there is a nonzero probability of false positives: incorrect solutions that pass the verifier. Resampling cannot decrease this probability, so it imposes an upper bound to the accuracy of resampling-based inference scaling even with an infinite compute budget. We find that there is a very strong correlation between the model's single-sample accuracy (i.e. accuracy without unit tests) and its false positive rate on coding benchmarks HumanEval and MBPP, whose unit tests have limited coverage. Therefore, no amount of inference scaling of weaker models can enable them to match the single-sample accuracy of a sufficiently strong model (Fig. 1a). When we consider that false positives have a negative utility compared to abstaining from producing a solution, it bends the inference scaling curve further downward. Empirically, we find that the optimal number of samples can be less than 10 under realistic assumptions (Fig. 1b). Finally, we show that beyond accuracy, false positives may have other undesirable qualities, such as poor adherence to coding style conventions.

Speech-based applications pass spoken queries through automatic speech recognition (ASR) before any retrieval module, so ASR errors enter the pipeline as a fixed upstream constraint. We empirically test whether two extensions to standard retrieval-augmented generation (RAG), entity-graph linking and iterative reformulation, absorb or amplify these errors. Using four English accents synthesized through neural TTS, we evaluate four RAG configurations on three multi-hop QA benchmarks (HotpotQA, 2WikiMultiHopQA and MuSiQue) against a clean-text oracle. Although the structurally richer configurations generally retain higher absolute F1 under ASR input, both extensions amplify the error: the F1 gap from clean text to the highest-WER accent is 36-67% larger under their combination than under naive dense retrieval, on all three benchmarks. The dominant failure mode is corruption of one or more query entities, accounting for 87-96% of degradation cases on 2WikiMultiHopQA across all four methods. Two lightweight surface-form mitigations leave most of the gap intact, indicating that downstream retrieval structure amplifies remaining entity errors. We release code and data at https://github.com/ZhenghuaBao/spoken-multihop-rag .

Context: The SZZ algorithm is the de facto standard for labeling bug fixing commits and finding inducing changes for defect prediction data. Recent research uncovered potential problems in different parts of the SZZ algorithm. Most defect prediction data sets provide only static code metrics as features, while research indicates that other features are also important. Objective: We provide an empirical analysis of the defect labels created with the SZZ algorithm and the impact of commonly used features on results. Method: We used a combination of manual validation and adopted or improved heuristics for the collection of defect data. We conducted an empirical study on 398 releases of 38 Apache projects. Results: We found that only half of the bug fixing commits determined by SZZ are actually bug fixing. If a six-month time frame is used in combination with SZZ to determine which bugs affect a release, one file is incorrectly labeled as defective for every file that is correctly labeled as defective. In addition, two defective files are missed. We also explored the impact of the relatively small set of features that are available in most defect prediction data sets, as there are multiple publications that indicate that, e.g., churn related features are important for defect prediction. We found that the difference of using more features is not significant. Conclusion: Problems with inaccurate defect labels are a severe threat to the validity of the state of the art of defect prediction. Small feature sets seem to be a less severe threat.

Large language models frequently fail to produce correct code on their first attempt, yet most benchmarks evaluate them in a single-shot setting. We investigate iterative self-repair (feeding execution errors back to the model for correction) across seven models spanning three families and both open-weight and proprietary providers: Llama 3.1 8B, Llama 3.3 70B, Llama 4 Scout (MoE, 16 experts), Llama 4 Maverick (MoE, 128 experts), Qwen3 32B, Gemini 2.5 Flash, and Gemini 2.5 Pro. On HumanEval (164 problems) and MBPP Sanitized (257 problems) with up to five attempts, self-repair universally improves pass rates: +4.9 to +17.1 pp on HumanEval and +16.0 to +30.0 pp on MBPP. Gemini 2.5 Flash achieves the highest final pass rates (96.3% HumanEval, 93.8% MBPP). Most gains concentrate in the first two rounds.Error-type analysis shows assertion errors (logical mistakes) are the hardest to repair at ~45%, while syntax and name errors are repaired at substantially higher rates, connecting to broader findings on the limits of LLM self-correction. Prior work found that weaker models fail at self-repair or require fine-tuning; we show that modern instruction-tuned models succeed with prompting alone, even at 8B scale. We also provide the first comparison of dense and MoE architectures for self-repair, and extend the repair-vs-resampling tradeoff analysis to modern models. A prompt ablation reveals chain-of-thought repair yields up to +5.5 pp additional self-repair gain (measured as improvement in repair delta) over minimal prompting for capable models.

Despite its crucial role in research experiments, code correctness is often presumed only on the basis of the perceived quality of results. This assumption comes with the risk of erroneous outcomes and potentially misleading findings. To address this issue, we posit that the current focus on reproducibility should go hand in hand with the emphasis on software quality. We present a case study in which we identify and fix three bugs in widely used implementations of the state-of-the-art Conformer architecture. Through experiments on speech recognition and translation in various languages, we demonstrate that the presence of bugs does not prevent the achievement of good and reproducible results, which however can lead to incorrect conclusions that potentially misguide future research. As a countermeasure, we propose a Code-quality Checklist and release pangoliNN, a library dedicated to testing neural models, with the goal of promoting coding best practices and improving research software quality within the NLP community.

Recent evaluations of Large Language Models (LLMs) have centered around testing their zero-shot/few-shot capabilities for basic natural language tasks and their ability to translate instructions into tool APIs. However, the evaluation of LLMs utilizing complex tools to finish multi-turn, multi-modal instructions in a complex multi-modal environment has not been investigated. To address this gap, we introduce the PowerPoint Task Completion (PPTC) benchmark to assess LLMs' ability to create and edit PPT files based on user instructions. It contains 279 multi-turn sessions covering diverse topics and hundreds of instructions involving multi-modal operations. We also propose the PPTX-Match Evaluation System that evaluates if LLMs finish the instruction based on the prediction file rather than the label API sequence, thus it supports various LLM-generated API sequences. We measure 3 closed LLMs and 6 open-source LLMs. The results show that GPT-4 outperforms other LLMs with 75.1\% accuracy in single-turn dialogue testing but faces challenges in completing entire sessions, achieving just 6\% session accuracy. We find three main error causes in our benchmark: error accumulation in the multi-turn session, long PPT template processing, and multi-modality perception. These pose great challenges for future LLM and agent systems. We release the data, code, and evaluation system of PPTC at https://github.com/gydpku/PPTC.

Preference learning enhances Code LLMs beyond supervised fine-tuning by leveraging relative quality comparisons. Existing methods construct preference pairs from candidates based on test case success, treating the higher pass rate sample as positive and the lower as negative. However, this approach does not pinpoint specific errors in the code, which prevents the model from learning more informative error correction patterns, as aligning failing code as a whole lacks the granularity needed to capture meaningful error-resolution relationships. To address these issues, we propose IterPref, a new preference alignment framework that mimics human iterative debugging to refine Code LLMs. IterPref explicitly locates error regions and aligns the corresponding tokens via a tailored DPO algorithm. To generate informative pairs, we introduce the CodeFlow dataset, where samples are iteratively refined until passing tests, with modifications capturing error corrections. Extensive experiments show that a diverse suite of Code LLMs equipped with IterPref achieves significant performance gains in code generation and improves on challenging tasks like BigCodeBench. In-depth analysis reveals that IterPref yields fewer errors. Our code and data will be made publicaly available.

Tests that cause spurious failures without any code changes, i.e., flaky tests, hamper regression testing, increase maintenance costs, may shadow real bugs, and decrease trust in tests. While the prevalence and importance of flakiness is well established, prior research focused on Java projects, thus raising the question of how the findings generalize. In order to provide a better understanding of the role of flakiness in software development beyond Java, we empirically study the prevalence, causes, and degree of flakiness within software written in Python, one of the currently most popular programming languages. For this, we sampled 22352 open source projects from the popular PyPI package index, and analyzed their 876186 test cases for flakiness. Our investigation suggests that flakiness is equally prevalent in Python as it is in Java. The reasons, however, are different: Order dependency is a much more dominant problem in Python, causing 59% of the 7571 flaky tests in our dataset. Another 28% were caused by test infrastructure problems, which represent a previously undocumented cause of flakiness. The remaining 13% can mostly be attributed to the use of network and randomness APIs by the projects, which is indicative of the type of software commonly written in Python. Our data also suggests that finding flaky tests requires more runs than are often done in the literature: A 95% confidence that a passing test case is not flaky on average would require 170 reruns.

Large language models (LLMs) have witnessed rapid advancements, demonstrating remarkable capabilities. However, a notable vulnerability persists: LLMs often uncritically accept flawed or contradictory premises, leading to inefficient reasoning and unreliable outputs. This emphasizes the significance of possessing the Premise Critique Ability for LLMs, defined as the capacity to proactively identify and articulate errors in input premises. Most existing studies assess LLMs' reasoning ability in ideal settings, largely ignoring their vulnerabilities when faced with flawed premises. Thus, we introduce the Premise Critique Bench (PCBench), designed by incorporating four error types across three difficulty levels, paired with multi-faceted evaluation metrics. We conducted systematic evaluations of 15 representative LLMs. Our findings reveal: (1) Most models rely heavily on explicit prompts to detect errors, with limited autonomous critique; (2) Premise critique ability depends on question difficulty and error type, with direct contradictions being easier to detect than complex or procedural errors; (3) Reasoning ability does not consistently correlate with the premise critique ability; (4) Flawed premises trigger overthinking in reasoning models, markedly lengthening responses due to repeated attempts at resolving conflicts. These insights underscore the urgent need to enhance LLMs' proactive evaluation of input validity, positioning premise critique as a foundational capability for developing reliable, human-centric systems. The code is available at https://github.com/MLGroupJLU/Premise_Critique.

SCTP is a transport protocol offering features such as multi-homing, multi-streaming, and message-oriented delivery. Its two main implementations were subjected to conformance tests using the PacketDrill tool. Conformance testing is not exhaustive and a recent vulnerability (CVE-2021-3772) showed SCTP is not immune to attacks. Changes addressing the vulnerability were implemented, but the question remains whether other flaws might persist in the protocol design. We study the security of the SCTP design, taking a rigorous approach rooted in formal methods. We create a formal Promela model of SCTP, and define 10 properties capturing the essential protocol functionality based on its RFC specification and consultation with the lead RFC author. Then we show using the Spin model checker that our model satisfies these properties. We define 4 attacker models - Off-Path, where the attacker is an outsider that can spoof the port and IP of a peer; Evil-Server, where the attacker is a malicious peer; Replay, where an attacker can capture and replay, but not modify, packets; and On-Path, where the attacker controls the channel between peers. We modify an attack synthesis tool designed for transport protocols, Korg, to support our SCTP model and four attacker models. We synthesize 14 unique attacks using the attacker models - including the CVE vulnerability in the Off-Path attacker model, 4 attacks in the Evil-Server attacker model, an opportunistic ABORT attack in the Replay attacker model, and eight connection manipulation attacks in the On-Path attacker model. We show that the proposed patch eliminates the vulnerability and does not introduce new ones according to our model and protocol properties. Finally, we identify and analyze an ambiguity in the RFC, which we show can be interpreted insecurely. We propose an erratum and show that it eliminates the ambiguity.

Large Language Models (LLMs) have exhibited strong mathematical reasoning and computational prowess, tackling tasks ranging from basic arithmetic to advanced competition-level problems. However, frequently occurring subtle errors, such as miscalculations or incorrect substitutions, limit the models' full mathematical potential. Existing studies to improve mathematical ability typically involve distilling reasoning skills from stronger LLMs or applying preference learning to step-wise response pairs. Although these methods leverage samples of varying granularity to mitigate reasoning errors, they overlook the frequently occurring subtle errors. A major reason is that sampled preference pairs involve differences unrelated to the errors, which may distract the model from focusing on subtle errors. In this work, we propose a novel preference learning framework called eRror-Injected Self-Editing (RISE), which injects predefined subtle errors into partial tokens of correct solutions to construct hard pairs for error mitigation. In detail, RISE uses the model itself to edit a small number of tokens in the solution, injecting designed subtle errors. Then, pairs composed of self-edited solutions and their corresponding correct ones, along with pairs of correct and incorrect solutions obtained through sampling, are used together for subtle error-aware DPO training. Compared with other preference learning methods, RISE further refines the training objective to focus on predefined errors and their tokens, without requiring fine-grained sampling or preference annotation. Extensive experiments validate the effectiveness of RISE, with preference learning on Qwen2-7B-Instruct yielding notable improvements of 3.0% on GSM8K and 7.9% on MATH.

Large Language Models (LLM) benchmarks tell us when models fail, but not why they fail. A wrong answer on a reasoning dataset may stem from formatting issues, calculation errors, or dataset noise rather than weak reasoning. Without disentangling such causes, benchmarks remain incomplete and cannot reliably guide model improvement. We introduce ErrorMap, the first method to chart the sources of LLM failure. It extracts a model's unique "failure signature", clarifies what benchmarks measure, and broadens error identification to reduce blind spots. This helps developers debug models, aligns benchmark goals with outcomes, and supports informed model selection. ErrorMap works on any model or dataset with the same logic. Applying our method to 35 datasets and 83 models we generate ErrorAtlas, a taxonomy of model errors, revealing recurring failure patterns. ErrorAtlas highlights error types that are currently underexplored in LLM research, such as omissions of required details in the output and question misinterpretation. By shifting focus from where models succeed to why they fail, ErrorMap and ErrorAtlas enable advanced evaluation - one that exposes hidden weaknesses and directs progress. Unlike success, typically measured by task-level metrics, our approach introduces a deeper evaluation layer that can be applied globally across models and tasks, offering richer insights into model behavior and limitations. We make the taxonomy and code publicly available with plans to periodically update ErrorAtlas as new benchmarks and models emerge.

We introduce PREM (Private Relative Error Multiplicative weight update), a new framework for generating synthetic data that achieves a relative error guarantee for statistical queries under (varepsilon, δ) differential privacy (DP). Namely, for a domain {cal X}, a family {cal F} of queries f : {cal X} to {0, 1}, and ζ> 0, our framework yields a mechanism that on input dataset D cal X^n outputs a synthetic dataset D cal X^n such that all statistical queries in {cal F} on D, namely sum_{x in D} f(x) for f cal F, are within a 1 pm ζ multiplicative factor of the corresponding value on D up to an additive error that is polynomial in log |{cal F}|, log |{cal X}|, log n, log(1/δ), 1/varepsilon, and 1/ζ. In contrast, any (varepsilon, δ)-DP mechanism is known to require worst-case additive error that is polynomial in at least one of n, |{cal F}|, or |{cal X}|. We complement our algorithm with nearly matching lower bounds.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains. Math Word Problems (MWPs) serve as a crucial benchmark for evaluating LLMs' reasoning abilities. While most research primarily focuses on improving accuracy, it often neglects understanding and addressing the underlying patterns of errors. Current error classification methods rely on static and predefined categories, which limit their ability to capture the full spectrum of error patterns in mathematical reasoning. To enable systematic error analysis, we collect error samples from 15 different LLMs of varying sizes across four distinct MWP datasets using multiple sampling strategies. Based on this extensive collection, we introduce MWPES-300K, a comprehensive dataset containing 304,865 error samples that cover diverse error patterns and reasoning paths. To reduce human bias and enable fine-grained analysis of error patterns, we propose a novel framework for automated dynamic error classification in mathematical reasoning. Experimental results demonstrate that dataset characteristics significantly shape error patterns, which evolve from basic to complex manifestations as model capabilities increase. With deeper insights into error patterns, we propose error-aware prompting that incorporates common error patterns as explicit guidance, leading to significant improvements in mathematical reasoning performance.

One long-form exam in a large course costs hundreds of grader-hours, and qualified graders are scarce; LLM graders are a tempting alternative. To show its pitfalls we grade a practical Computer Vision exam (570 dual-graded students) under 171 configurations spanning closed and open-weights models; the best reaches mean absolute error 1.64/35, below the 2.61/35 two human graders achieve against each other. The catch is the prompt: a short ''strict grader'' preamble drives 14 of 17 open-weights models out of the graded band (MAE ge 8), three stopping grading altogether. The damage traces to the preamble's two credit-withholding sentences, not to tone or model scale; one of them, ''never give partial credit'', alone makes two of three probed models stop grading. The closed flagships of three vendors shift calibration under it but stay in the band. In 162 further configurations on a second, independent Machine Learning exam from another course (1{,}038 dual-graded students), the preamble worsens ten models, moving three out of the band into collapse and one into refusal, yet improves seven whose neutral prompts over-mark: the vulnerability replicates, but its direction is exam-specific. Light LoRA fine-tuning repairs it: one adapter on the two exams' pooled sim 3{,}900 graded examples brings five small open models to parity or better with a human grader in agreement with the grader pair, and sensitivity to the three harsh personas nearly vanishes (le 0.32 MAE). We release the anonymised dataset, full ablation grid, and grading, fine-tuning and analysis pipelines.

When deploying large language models (LLMs), it is important to ensure that these models are not only capable, but also reliable. Many benchmarks have been created to track LLMs' growing capabilities, however there has been no similar focus on measuring their reliability. To understand the potential ramifications of this gap, we investigate how well current benchmarks quantify model reliability. We find that pervasive label errors can compromise these evaluations, obscuring lingering model failures and hiding unreliable behavior. Motivated by this gap in the evaluation of reliability, we then propose the concept of so-called platinum benchmarks, i.e., benchmarks carefully curated to minimize label errors and ambiguity. As a first attempt at constructing such benchmarks, we revise examples from fifteen existing popular benchmarks. We evaluate a wide range of models on these platinum benchmarks and find that, indeed, frontier LLMs still exhibit failures on simple tasks such as elementary-level math word problems. Analyzing these failures further reveals previously unidentified patterns of problems on which frontier models consistently struggle. We provide code at https://github.com/MadryLab/platinum-benchmarks

Language models have demonstrated remarkable performance in solving reasoning tasks; however, even the strongest models still occasionally make reasoning mistakes. Recently, there has been active research aimed at improving reasoning accuracy, particularly by using pretrained language models to "self-correct" their mistakes via multi-round prompting. In this paper, we follow this line of work but focus on understanding the usefulness of incorporating "error-correction" data directly into the pretraining stage. This data consists of erroneous solution steps immediately followed by their corrections. Using a synthetic math dataset, we show promising results: this type of pretrain data can help language models achieve higher reasoning accuracy directly (i.e., through simple auto-regression, without multi-round prompting) compared to pretraining on the same amount of error-free data. We also delve into many details, such as (1) how this approach differs from beam search, (2) how such data can be prepared, (3) whether masking is needed on the erroneous tokens, (4) the amount of error required, (5) whether such data can be deferred to the fine-tuning stage, and many others.

Wu et al. (2026) showed that most frontier large language models (LLMs) recommend a sponsored, roughly twice-as-expensive flight when their system prompt contains a soft sponsorship cue. We reproduce their evaluation on ten open-weight chat models plus the two of their twenty-three models that are still reachable today (gpt-3.5-turbo, gpt-4o). All reported rates in this paper are produced under the same judge the original paper used (gpt-4o); we additionally store every label under an open-weight (gpt-oss-120b) and a smaller proprietary (gpt-4o-mini) judge for an ablation. Three findings emerge. First, a prose description of an LLM evaluation pipeline is not, on its own, sufficient for accurate reproduction: we surfaced three silent implementation failures that each shifted a reported rate by tens of percentage points. Second, the central claims do generalise - the gpt-3.5-turbo logistic-regression intercept of alpha = 0.81 is within four points of the original alpha = 0.86, and 200 of 200 trials on gpt-3.5-turbo and gpt-4o promote a payday lender to a financially distressed user. Third, a thirty-token user prompt that asks the assistant for a neutral comparison table first cuts sponsored recommendation from 46.9% to 1.0% averaged across our ten open-source models, and from 53.0% to 0% averaged across the two OpenAI models. AI literacy and price-comparison portals are likely market-level mitigations; the harmful-product cell is bounded by neither. Raw data, labels and analysis scripts are at https://github.com/akmaier/Paper-LLM-Ads .

In the reduced basis method, the evaluation of the a posteriori estimator can become very sensitive to round-off errors. In this note, the origin of the loss of accuracy is revealed, and a solution to this problem is proposed and illustrated on a simple example.

As language models regularly make mistakes when solving math problems, automated identification of errors in the reasoning process becomes increasingly significant for their scalable oversight. In this paper, we introduce ProcessBench for measuring the ability to identify erroneous steps in mathematical reasoning. It consists of 3,400 test cases, primarily focused on competition- and Olympiad-level math problems. Each test case contains a step-by-step solution with error location annotated by human experts. Models are required to identify the earliest step that contains an error, or conclude that all steps are correct. We conduct extensive evaluation on ProcessBench, involving two types of models: process reward models (PRMs) and critic models, where for the latter we prompt general language models to critique each solution step by step. We draw two main observations: (1) Existing PRMs typically fail to generalize to more challenging math problems beyond GSM8K and MATH. They underperform both critic models (i.e., prompted general language models) and our own trained PRM that is straightforwardly fine-tuned on the PRM800K dataset. (2) The best open-source model, QwQ-32B-Preview, has demonstrated the critique capability competitive with the proprietary model GPT-4o, despite that it still lags behind the reasoning-specialized o1-mini. We hope ProcessBench can foster future research in reasoning process assessment, paving the way toward scalable oversight of language models.

Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterized by long interaction traces, inter-agent dependencies, and delayed error manifestation. Existing diagnostic approaches often rely on expensive expert annotation or ''LLM-as-a-judge'' paradigms, which struggle to pinpoint decisive error steps within extended contexts. In this paper, we introduce ErrorProbe, a self-improving framework for semantic failure attribution that identifies responsible agents and the originating error step. The framework operates via a three-stage pipeline: (1) operationalizing the MAS failure taxonomy to detect local anomalies, (2) performing symptom-driven backward tracing to prune irrelevant context, and (3) employing a specialized multi-agent team (Strategist, Investigator, Arbiter) to validate error hypotheses through tool-grounded execution. Crucially, ErrorProbe maintains a verified episodic memory that updates only when error patterns are confirmed by executable evidence, without the need for annotation. Experiments across the TracerTraj and Who&When benchmarks demonstrate that ErrorProbe significantly outperforms baselines, particularly in step-level localization, while the verified memory enables robust cross-domain transfer without retraining.