Recent advancements in reinforcement learning (RL) for large language models (LLMs), exemplified by DeepSeek R1, have shown that even a simple question-answering task can substantially improve an LLM's reasoning capabilities. In this work, we extend this approach by modifying the task into a multi-attempt setting. Instead of generating a single response per question, the model is given multiple attempts, with feedback provided after incorrect responses. The multi-attempt task encourages the model to refine its previous attempts and improve search efficiency. Experimental results show that even a small LLM trained on a multi-attempt task achieves significantly higher accuracy when evaluated with more attempts, improving from 45.6% with 1 attempt to 52.5% with 2 attempts on the math benchmark. In contrast, the same LLM trained on a standard single-turn task exhibits only a marginal improvement, increasing from 42.3% to 43.2% when given more attempts during evaluation. The results indicate that, compared to the standard single-turn task, an LLM trained on a multi-attempt task achieves slightly better performance on math benchmarks while also learning to refine its responses more effec
State-of-the-art reasoning models utilize long chain-of-thought (CoT) to solve increasingly complex problems using more test-time computation. In this work, we explore a long CoT setting where the model makes up to K successive attempts at solving a problem, in which each attempt is allowed to build on earlier ones after the model receives a hard verifier feedback. This motivates RL methods that can harness per-attempt rewards by carefully weighting individual attempts. We study optimizing the Verification@K reward (the model succeeds by the K-th attempt) and show that naively weighing the attempts by their pass/fail results in biased gradients. We introduce Calibrated Attempt-Level (CAL) GRPO by devising a weighing strategy to obtain unbiased gradients while maintaining small variance. Our theory reveals how incorporating per-attempt rewards influence the training and the eventual Verification@K performance. Experiments, baselines, and ablations on synthetic and real data corroborate our theory and the benefits of CAL-GRPO over vanilla GRPO as well as naive weighting.
The current paper examines the possibility of replacing conventional synchronous single-attempt exam with more flexible and accessible multi-attempt asynchronous assessments in introductory-level physics by using large isomorphic problem banks. We compared student's performance on both numeric and conceptual problems administered on a multi-attempt, asynchronous quiz to their performance on isomorphic problems administered on a subsequent single-attempt, synchronous exam. We computed the phi coefficient and the McNemar's test statistic for the correlation matrix between paired problems on both assessments as a function of the number of attempts considered on the quiz. We found that for the conceptual problems, a multi-attempt quiz with five allowed attempts could potentially replace similar problems on a single-attempt exam, while there was a much weaker association for the numerical questions beyond two quiz attempts.
Large language models (LLMs) have achieved substantial progress in repository-level code generation. However, solving the same repository-level task often requires multiple attempts, while existing methods still optimize each attempt in isolation and do not preserve or reuse task-specific state across attempts. In this paper, we propose LiveCoder, a novel framework for repository-level code generation based on cross-attempt knowledge optimization. LiveCoder maintains persistent task-specific state from prior attempts to guide subsequent generation. This state includes success knowledge, which captures reusable signals from previously strong repositories, failure knowledge, which records unsuccessful outcomes and their diagnostic signals, and a historical-best repository, which preserves the strongest result found so far and prevents regression. These components collectively transform repeated repository generation into a persistent, knowledge-driven optimization process. We evaluate LiveCoder using four frontier LLMs on two representative repository-level code generation benchmarks. Extensive experimental results demonstrate the effectiveness and efficiency of LiveCoder, improving
On July 13, 2024, at the Trump rally in Pennsylvania, someone attempted to assassinate Republican Presidential Candidate Donald Trump. This attempt sparked a large-scale discussion on social media. We collected posts from X (formerly known as Twitter) one week before and after the assassination attempt and aimed to model the short-term effects of such a ``shock'' on public opinions and discussion topics. Specifically, our study addresses three key questions: first, we investigate how public sentiment toward Donald Trump shifts over time and across regions (RQ1) and examine whether the assassination attempt itself significantly affects public attitudes, independent of the existing political alignments (RQ2). Finally, we explore the major themes in online conversations before and after the crisis, illustrating how discussion topics evolved in response to this politically charged event (RQ3). By integrating large language model-based sentiment analysis, difference-in-differences modeling, and topic modeling techniques, we find that following the attempt the public response was broadly sympathetic to Trump rather than polarizing, despite baseline ideological and regional disparities.
Persuasion is a powerful capability of large language models (LLMs) that both enables beneficial applications (e.g. helping people quit smoking) and raises significant risks (e.g. large-scale, targeted political manipulation). Prior work has found models possess a significant and growing persuasive capability, measured by belief changes in simulated or real users. However, these benchmarks overlook a crucial risk factor: the propensity of a model to attempt to persuade in harmful contexts. Understanding whether a model will blindly ``follow orders'' to persuade on harmful topics (e.g. glorifying joining a terrorist group) is key to understanding the efficacy of safety guardrails. Moreover, understanding if and when a model will engage in persuasive behavior in pursuit of some goal is essential to understanding the risks from agentic AI systems. We propose the Attempt to Persuade Eval (APE) benchmark, that shifts the focus from persuasion success to persuasion attempts, operationalized as a model's willingness to generate content aimed at shaping beliefs or behavior. Our evaluation framework probes frontier LLMs using a multi-turn conversational setup between simulated persuader and
This paper investigates several distinct attempts to generalize in higher dimension the standard 2-dimensional phyllotaxy set construction. We first recall known contructions for these sets on $2D$ manifolds of constant curvature (the Euclidean plane $\mathbb{R}^2$, the sphere $\mathbb{S}^2$ and the hyperbolic plane $\mathbb{H}^2$). We then propose a first attempt to get a $3D$ phyllotactic set by piling up suitably shifted Euclidean $2D$ phyllotactic sets. A different, radially triggered, solution is then analyzed. An interesting phyllotactic set on the hypersphere $\mathbb{S}^3$ is then generated using a Hopf fibration approach. Finally,a simple 4-dimensional example is presented, generated as a simple product of two 2-dimensional planar sets. A $3D$ phyllotaxy candidate is then derived by applying a "Cut and Project" algorithm.
In continuous-variable quantum key distribution (CV-QKD), the performance of the information reconciliation (IR) step is critical for the achievable secret key rate (SKR) and transmission distance. We show how to improve on the recently introduced implementation of an IR-protocol involving multiple decoding attempts (MDA) and validate the method on simulated data in different application scenarios. Throughout, we demonstrate meaningful SKR-gains compared to both the standard protocol of a single decoding attempt and to the original MDA-implementation, even at given decoding complexity.
Photonic interconnects are a key technology for scaling up atomic based quantum computers. By facilitating the connection of multiple systems, high-performance modular quantum processing units may be constructed to perform deeper and more useful algorithms. Most previous implementations of photonic interconnects in trapped ions utilize the scheme of preparing a state, exciting it, and collecting single photons from decays of the excited state. State preparation is responsible for the vast majority of the total attempt time, often taking hundreds of nanoseconds to several microseconds. Here, we describe and analyze a novel technique called ``electron juggling" to speed up photonic interconnects by reducing the state preparation step substantially. Using a theoretical framework, we illustrate how this scheme can significantly increase remote entanglement generation rates, approaching the atomic physics limit of the attempt rate in trapped-ion photonic interconnects. Our results indicate that this scheme holds the possibility of achieving remote entanglement generation rates of over 1,000 Bell pairs per second.
Despite a surge in robotics research dedicated to inferring and understanding human intent, a universally accepted definition remains elusive since existing works often equate human intention with specific task-related goals. This article seeks to address this gap by examining the multifaceted nature of intention. Drawing on insights from psychology, it attempts to consolidate a definition of intention into a comprehensible framework for a broader audience. The article classifies different types of intention based on psychological and communication studies, offering guidance to researchers shifting from pure technical enhancements to a more human-centric perspective in robotics. It then demonstrates how various robotics studies can be aligned with these intention categories. Finally, through in-depth analyses of collaborative search and object transport use cases, the article underscores the significance of considering the diverse facets of human intention.
The Arrhenius law predicts the transition time between equilibrium states in physical systems due to thermal activation, with broad applications in material science, magnetic hyperthermia and paleomagnetism where it is used to estimate the transition time and thermal stability of assemblies of magnetic nanoparticles. Magnetite is a material of great importance in paleomagnetic studies and magnetic hyperthermia but existing estimates of the attempt frequency $f_0$ vary by several orders of magnitude in the range $10^7-10^{13}$ Hz, leading to significant uncertainty in their relaxation rate. Here we present a dynamical method enabling full parameterization of the Arrhenius-Néel law using atomistic spin dynamics. We determine the temperature and volume dependence of the attempt frequency of magnetite nanoparticles with cubic anisotropy and find a value of $f_0 = 0.562 \pm 0.059$ GHz at room temperature. For particles with enhanced anisotropy we find a significant increase in the attempt frequency and a strong temperature dependence suggesting an important role of anisotropy. The method is applicable to a wide range of dynamical systems where different states can be clearly identified
In longitudinal studies, it is not uncommon to make multiple attempts to collect a measurement after baseline. Recording whether these attempts are successful provides useful information for the purposes of assessing missing data assumptions. This is because measurements from subjects who provide the data after numerous failed attempts may differ from those who provide the measurement after fewer attempts. Previous models for these designs were parametric and/or did not allow sensitivity analysis. For the former, there are always concerns about model misspecification and for the latter, sensitivity analysis is essential when conducting inference in the presence of missing data. Here, we propose a new approach which minimizes issues with model misspecification by using Bayesian nonparametrics for the observed data distribution. We also introduce a novel approach for identification and sensitivity analysis. We re-analyze the repeated attempts data from a clinical trial involving patients with severe mental illness and conduct simulations to better understand the properties of our approach.
Hoffmann et al. (2022) propose three methods for estimating a compute-optimal scaling law. We attempt to replicate their third estimation procedure, which involves fitting a parametric loss function to a reconstruction of data from their plots. We find that the reported estimates are inconsistent with their first two estimation methods, fail at fitting the extracted data, and report implausibly narrow confidence intervals--intervals this narrow would require over 600,000 experiments, while they likely only ran fewer than 500. In contrast, our rederivation of the scaling law using the third approach yields results that are compatible with the findings from the first two estimation procedures described by Hoffmann et al.
Recently Shekhar Suman [arXiv: 2407.07121v6 [math.GM] 3 Aug 2024] made an attempt to prove the irrationality of $ζ(5)$. But unfortunately the proof is not correct. In this note, we discuss the fallacy in the proof.
This technical report presents our initial attempt to build a spoken large language model (LLM) for Taiwanese Mandarin, specifically tailored to enable real-time, speech-to-speech interaction in multi-turn conversations. Our end-to-end model incorporates a decoder-only transformer architecture and aims to achieve seamless interaction while preserving the conversational flow, including full-duplex capabilities allowing simultaneous speaking and listening. The paper also details the training process, including data preparation with synthesized dialogues and adjustments for real-time interaction. We also developed a platform to evaluate conversational fluency and response coherence in multi-turn dialogues. We hope the release of the report can contribute to the future development of spoken LLMs in Taiwanese Mandarin.
Suicide is an important public health concern and one of the leading causes of death worldwide. Suicidal behaviors, including suicide attempts (SA) and suicide ideations (SI), are leading risk factors for death by suicide. Information related to patients' previous and current SA and SI are frequently documented in the electronic health record (EHR) notes. Accurate detection of such documentation may help improve surveillance and predictions of patients' suicidal behaviors and alert medical professionals for suicide prevention efforts. In this study, we first built Suicide Attempt and Ideation Events (ScAN) dataset, a subset of the publicly available MIMIC III dataset spanning over 12k+ EHR notes with 19k+ annotated SA and SI events information. The annotations also contain attributes such as method of suicide attempt. We also provide a strong baseline model ScANER (Suicide Attempt and Ideation Events Retriever), a multi-task RoBERTa-based model with a retrieval module to extract all the relevant suicidal behavioral evidences from EHR notes of an hospital-stay and, and a prediction module to identify the type of suicidal behavior (SA and SI) concluded during the patient's stay at th
We consider the problem of minimizing a differentiable function with locally Lipschitz continuous gradient over the real determinantal variety, and present a first-order algorithm designed to find stationary points of that problem. This algorithm applies steps of a retraction-free descent method proposed by Schneider and Uschmajew (2015), while taking the numerical rank into account to attempt rank reductions. We prove that this algorithm produces a sequence of iterates the accumulation points of which are stationary, and therefore does not follow the so-called apocalypses described by Levin, Kileel, and Boumal (2022). Moreover, the rank reduction mechanism of this algorithm requires at most one rank reduction attempt per iteration, in contrast with the one of the $\mathrm{P}^2\mathrm{GDR}$ algorithm introduced by Olikier, Gallivan, and Absil (2022) which can require a number of rank reduction attempts equal to the rank of the iterate in the worst-case scenario.
Many applications of formal methods require automated reasoning about system properties, such as system safety and security. To improve the performance of automated reasoning engines, such as SAT/SMT solvers and first-order theorem prover, it is necessary to understand both the successful and failing attempts of these engines towards producing formal certificates, such as logical proofs and/or models. Such an analysis is challenging due to the large number of logical formulas generated during proof/model search. In this paper we focus on saturation-based first-order theorem proving and introduce the SATVIS tool for interactively visualizing saturation-based proof attempts in first-order theorem proving. We build SATVIS on top of the world-leading theorem prover VAMPIRE, by interactively visualizing the saturation attempts of VAMPIRE in SATVIS. Our work combines the automatic layout and visualization of the derivation graph induced by the saturation attempt with interactive transformations and search functionality. As a result, we are able to analyze and debug (failed) proof attempts of VAMPIRE. Thanks to its interactive visualisation, we believe SATVIS helps both experts and non-ex
This work introduces a new multi-task, parameter-efficient language model (LM) tuning method that learns to transfer knowledge across different tasks via a mixture of soft prompts-small prefix embedding vectors pre-trained for different tasks. Our method, called ATTEMPT (ATTEntional Mixtures of Prompt Tuning), obtains source prompts as encodings of large-scale source tasks into a small number of parameters and trains an attention module to interpolate the source prompts and a newly initialized target prompt for every instance in the target task. During training, only the target task prompt and the attention weights, which are shared between tasks in multi-task training, are updated, while the original LM and source prompts are intact. ATTEMPT is highly parameter-efficient (e.g., updates 2,300 times fewer parameters than full fine-tuning) while achieving high task performance using knowledge from high-resource tasks. Moreover, it is modular using pre-trained soft prompts, and can flexibly add or remove source prompts for effective knowledge transfer. Our experimental results across 21 diverse NLP datasets show that ATTEMPT significantly outperforms prompt tuning and outperforms or m
Block-based visual programming environments are increasingly used to introduce computing concepts to beginners. Given that programming tasks are open-ended and conceptual, novice students often struggle when learning in these environments. AI-driven programming tutors hold great promise in automatically assisting struggling students, and need several components to realize this potential. We investigate the crucial component of student modeling, in particular, the ability to automatically infer students' misconceptions for predicting (synthesizing) their behavior. We introduce a novel benchmark, StudentSyn, centered around the following challenge: For a given student, synthesize the student's attempt on a new target task after observing the student's attempt on a fixed reference task. This challenge is akin to that of program synthesis; however, instead of synthesizing a {solution} (i.e., program an expert would write), the goal here is to synthesize a {student attempt} (i.e., program that a given student would write). We first show that human experts (TutorSS) can achieve high performance on the benchmark, whereas simple baselines perform poorly. Then, we develop two neuro/symbolic