Large-scale reinforcement learning with verifiable rewards (RLVR) has demonstrated its effectiveness in harnessing the potential of large language models (LLMs) for single-turn reasoning tasks. In realistic reasoning scenarios, LLMs can often utilize external tools to assist in task-solving processes. However, current RL algorithms inadequately balance the models' intrinsic long-horizon reasoning capabilities and their proficiency in multi-turn tool interactions. To bridge this gap, we propose Agentic Reinforced Policy Optimization (ARPO), a novel agentic RL algorithm tailored for training multi-turn LLM-based agents. Through preliminary experiments, we observe that LLMs tend to exhibit highly uncertain behavior, characterized by an increase in the entropy distribution of generated tokens, immediately following interactions with external tools. Motivated by this observation, ARPO incorporates an entropy-based adaptive rollout mechanism, dynamically balancing global trajectory sampling and step-level sampling, thereby promoting exploration at steps with high uncertainty after tool usage. By integrating an advantage attribution estimation, ARPO enables LLMs to internalize advantage differences in stepwise tool-use interactions. Our experiments across 13 challenging benchmarks in computational reasoning, knowledge reasoning, and deep search domains demonstrate ARPO's superiority over trajectory-level RL algorithms. Remarkably, ARPO achieves improved performance using only half of the tool-use budget required by existing methods, offering a scalable solution for aligning LLM-based agents with real-time dynamic environments. Our code and datasets are released at https://github.com/dongguanting/ARPO

Retrieval-Augmented Generation (RAG) models have drawn considerable attention in modern open-domain question answering. The effectiveness of RAG depends on the quality of the top retrieved documents. However, conventional retrieval methods sometimes fail to rank the most relevant documents at the top. In this paper, we introduce ASRank, a new re-ranking method based on scoring retrieved documents using zero-shot answer scent which relies on a pre-trained large language model to compute the likelihood of the document-derived answers aligning with the answer scent. Our approach demonstrates marked improvements across several datasets, including NQ, TriviaQA, WebQA, ArchivalQA, HotpotQA, and Entity Questions. Notably, ASRank increases Top-1 retrieval accuracy on NQ from $19.2\%$ to $46.5\%$ for MSS and $22.1\%$ to $47.3\%$ for BM25. It also shows strong retrieval performance on several datasets compared to state-of-the-art methods (47.3 Top-1 by ASRank vs 35.4 by UPR by BM25).

Knowledge probing evaluates the extent to which a language model (LM) has acquired relational knowledge during its pre-training phase. It provides a cost-effective means of comparing LMs of different sizes and training setups and is useful for monitoring knowledge gained or lost during continual learning (CL). In prior work, we presented an improved knowledge probe called BEAR (Wiland et al., 2024), which enables the comparison of LMs trained with different pre-training objectives (causal and masked LMs) and addresses issues of skewed distributions in previous probes to deliver a more unbiased reading of LM knowledge. With this paper, we present LM-PUB- QUIZ, a Python framework and leaderboard built around the BEAR probing mechanism that enables researchers and practitioners to apply it in their work. It provides options for standalone evaluation and direct integration into the widely-used training pipeline of the Hugging Face TRANSFORMERS library. Further, it provides a fine-grained analysis of different knowledge types to assist users in better understanding the knowledge in each evaluated LM. We publicly release LM-PUB-QUIZ as an open-source project.

Knowledge probing assesses to which degree a language model (LM) has successfully learned relational knowledge during pre-training. Probing is an inexpensive way to compare LMs of different sizes and training configurations. However, previous approaches rely on the objective function used in pre-training LMs and are thus applicable only to masked or causal LMs. As a result, comparing different types of LMs becomes impossible. To address this, we propose an approach that uses an LM's inherent ability to estimate the log-likelihood of any given textual statement. We carefully design an evaluation dataset of 7,731 instances (40,916 in a larger variant) from which we produce alternative statements for each relational fact, one of which is correct. We then evaluate whether an LM correctly assigns the highest log-likelihood to the correct statement. Our experimental evaluation of 22 common LMs shows that our proposed framework, BEAR, can effectively probe for knowledge across different LM types. We release the BEAR datasets and an open-source framework that implements the probing approach to the research community to facilitate the evaluation and development of LMs.

The biodiversity of our planet is under threat, with approximately one million species expected to become extinct within decades. The reason; negative human actions, which include hunting, overfishing, pollution, and the conversion of land for urbanisation and agricultural purposes. Despite significant investment from charities and governments for activities that benefit nature, global wildlife populations continue to decline. Local wildlife guardians have historically played a critical role in global conservation efforts and have shown their ability to achieve sustainability at various levels. In 2021, COP26 recognised their contributions and pledged US$1.7 billion per year; however, this is a fraction of the global biodiversity budget available (between US$124 billion and US$143 billion annually) given they protect 80% of the planets biodiversity. This paper proposes a radical new solution based on "Interspecies Money," where animals own their own money. Creating a digital twin for each species allows animals to dispense funds to their guardians for the services they provide. For example, a rhinoceros may release a payment to its guardian each time it is detected in a camera trap as long as it remains alive and well. To test the efficacy of this approach 27 camera traps were deployed over a 400km2 area in Welgevonden Game Reserve in Limpopo Province in South Africa. The motion-triggered camera traps were operational for ten months and, using deep learning, we managed to capture images of 12 distinct animal species. For each species, a makeshift bank account was set up and credited with {\pounds}100. Each time an animal was captured in a camera and successfully classified, 1 penny (an arbitrary amount - mechanisms still need to be developed to determine the real value of species) was transferred from the animal account to its associated guardian.

The rapidly increasing number of internet of things (IoT) and non-IoT devices has imposed new security challenges to network administrators. Accurate device identification in the increasingly complex network structures is necessary. In this paper, a device fingerprinting (DFP) method has been proposed for device identification, based on digital footprints, which devices use for communication over a network. A subset of nine features have been selected from the network and transport layers of a single transmission control protocol/internet protocol packet based on attribute evaluators in Weka, to generate device-specific signatures. The method has been evaluated on two online datasets, and an experimental dataset, using different supervised machine learning (ML) algorithms. Results have shown that the method is able to distinguish device type with up to 100% precision using the random forest (RF) classifier, and classify individual devices with up to 95.7% precision. These results demonstrate the applicability of the proposed DFP method for device identification, in order to provide a more secure and robust network.