国际商业机器公司

We present ThinkPrune, a simple yet effective method for pruning the thinking length for long-thinking LLMs, which has been found to often produce inefficient and redundant thinking processes. Existing preliminary explorations of reducing thinking length primarily focus on forcing the thinking process to early exit, rather than adapting the LLM to optimize and consolidate the thinking process, and therefore the length-performance tradeoff observed so far is sub-optimal. To fill this gap, ThinkPrune offers a simple solution that continuously trains the long-thinking LLMs via reinforcement learning (RL) with an added token limit, beyond which any unfinished thoughts and answers will be discarded, resulting in a zero reward. To further preserve model performance, we introduce an iterative length pruning approach, where multiple rounds of RL are conducted, each with an increasingly more stringent token limit. We observed that ThinkPrune results in a remarkable performance-length tradeoff on the AIME24 dataset, the reasoning length of DeepSeek-R1-Distill-Qwen-1.5B can be reduced by half with only 2% drop in performance. We also observed that after pruning, the LLMs can bypass unnecessary steps while keeping the core reasoning process complete.

The generation of highly accurate models of behaviours of individual cells and cell populations through integration of high-resolution assays with advanced computational tools would transform precision medicine. Recent breakthroughs in single-cell and spatial transcriptomics and multi-omics technologies, coupled with artificial intelligence, are driving rapid progress in model development. Complementing the advances in artificial intelligence, quantum computing is maturing as a novel compute paradigm that may offer potential solutions to overcome the computational bottlenecks inherent to capturing cellular dynamics. In this Roadmap article, we discuss the advancements and challenges in spatiotemporal single-cell analysis, explore the possibility of quantum computing to address the challenges and present a case study on how quantum computing may be integrated into cell-based therapeutics. The specific confluence of quantum and classical computing with high-resolution assays may offer a crucial path towards the generation of transformative models of cellular behaviours and perturbation responses. Classical computing has inherent limitations in capturing cellular dynamics. This Roadmap article discusses how recent advancements in quantum computing could overcome bottlenecks in spatiotemporal single-cell omics analyses and how it may be integrated into cell-based therapeutics.

Interorganizational ecosystems require governance arrangements that can align diverse and often competing organizations around a shared value proposition. Although blockchain, as a form of digital governance, promises to facilitate large-scale collaboration by codifying and enforcing rules, decentralizing control, and ensuring verifiable data exchange, many blockchain-based interorganizational ecosystems nevertheless fail. Leveraging 155 interviews and detailed internal records from a large technology provider covering 81 interorganizational blockchain projects across 25 industries, supplemented by archival evidence on the trajectories of 196 projects and 70 podcast interviews, we develop a grounded theory explaining how governance misalignments trigger collaboration breakdowns. Specifically, we identify three underlying governance tradeoffs that expose tensions between blockchain's network-centric rules and actor-centric needs: consistency versus flexibility in coordination, system reliance versus actor reliance in trust and control, and ecosystem utility versus member utility in incentives. These tradeoffs are amplified or attenuated by corresponding boundary conditions related to scale and cohesion (for coordination), co-opetition (for trust and control), and value logics (for incentives). Our study advances governance theory by explaining how blockchain interacts with traditional governance, shapes critical tradeoffs, and influences ecosystem success and failure. We also offer design principles to help managers navigate the inherent governance tradeoffs in ecosystem collaboration.

Long-term memory is a critical capability for multimodal large language model (MLLM) agents, particularly in conversational settings where information accumulates and evolves over time. However, existing benchmarks either evaluate multi-session memory in text-only conversations or assess multimodal understanding within localized contexts, failing to evaluate how multimodal memory is preserved, organized, and evolved across long-term conversational trajectories. Thus, we introduce Mem-Gallery, a new benchmark for evaluating multimodal long-term conversational memory in MLLM agents. Mem-Gallery features high-quality multi-session conversations grounded in both visual and textual information, with long interaction horizons and rich multimodal dependencies. Building on this dataset, we propose a systematic evaluation framework that assesses key memory capabilities along three functional dimensions: memory extraction and test-time adaptation, memory reasoning, and memory knowledge management. Extensive benchmarking across twelve memory systems reveals several key findings, highlighting the necessity of explicit multimodal information retention and memory organization, the persistent limitations in memory reasoning and knowledge management, as well as the efficiency bottleneck of current models. Our benchmark and dataset are available at https://github.com/YuanchenBei/Mem-Gallery.

Agent skills, which are reusable, domain-specific knowledge artifacts, have become a popular mechanism for extending LLM-based agents, yet formally benchmarking skill usage performance remains scarce. Existing skill benchmarking efforts focus primarily on idealized conditions, where LLMs are directly provided with hand-crafted, narrowly-tailored skills for each task. This setting reflects customized workflow automation, but does not cover the complementary case where the LLM agent must search for and select relevant skills on its own, and even the closest matching skills may not be well-tailored for the task. In this paper, we conduct the first comprehensive study of skill utility under progressively challenging realistic settings, where agents must retrieve skills from a large collection of 34k real-world skills and may not have access to any hand-curated skills. Our findings reveal that the benefits of skills are fragile: performance gains degrade consistently as settings become more realistic, with pass rates approaching no-skill baselines in the most challenging scenarios. To narrow this gap, we study skill refinement strategies, including query-specific and query-agnostic approaches, and we show that query-specific refinement substantially recovers lost performance when the initial skills are of reasonable relevance and quality. We further demonstrate the generality of retrieval and refinement on Terminal-Bench 2.0, where they improve the pass rate of Claude Opus 4.6 from 57.7% to 65.5%. Our results, consistent across multiple models, highlight both the promise and the current limitations of skills for LLM-based agents. Our code is available at https://github.com/UCSB-NLP-Chang/Skill-Usage.