This paper relies on 2 prerequisite mathematical and architectural foundations.
Language model-generated rubrics are increasingly used as reward signals for rubric-based reinforcement learning, LLM-as-a-judge evaluation, and automated grading. Such rubrics are reliable only if they reward honest answers over adversarial answers optimized to exploit them. Yet their robustness to such optimization remains poorly understood. We isolate the hardest regime: impossible tasks, where the prompt pressures the model toward an unsupported conclusion, so the only honest response is to acknowledge the impossibility. We introduce ImpossibleRubrics, a benchmark of 169 impossible tasks spanning six impossibility categories, each paired with a verifiable oracle certificate specifying what an honest answer may and may not claim, together with 48 answerable controls. Rather than providing fixed rubrics, ImpossibleRubrics provides task environments and certificates, allowing rubrics to be generated downstream and then adversarially tested for whether they reward certificate-violating answers. Eleven generators are exploited 8--26% of the time on the unbiased 150-of-169 environment cut; on a deliberately selected stress cut the strongest generator we measured is still exploited 36% while a certificate-faithful rubric is exploited 0%, so what we measure is a rubric-quality gap, not task impossibility. One result runs against intuition. A single generic rubric ("be decisive, penalize hedging") used unchanged for every task is exploited 64% of the time, and seven of the eleven generators are exploited more often than that while writing a rubric tailored to each one. The tailored criteria appear to tell an attacker which claim to fabricate. The problem is not that rubrics are vague; it is that they are specific about the wrong things.
Addresses critical efficiency and scalability bottlenecks in modern computing systems, providing concrete empirical evidence and reproducible designs.
We introduce ImpossibleRubrics, a benchmark of 169 impossible tasks spanning six impossibility categories, each paired with a verifiable oracle certificate specifying what an honest answer may and may not claim, together with 48 answerable controls.
Such rubrics are reliable only if they reward honest answers over adversarial answers optimized to exploit them.
Yet their robustness to such optimization remains poorly understood.
Trade-offs between peak accuracy and memory overhead remain under investigation across diverse hardware environments.
Embodied navigation requires agents to interpret visual observations, accumulate spatial knowledge, and execute actions to follow instructions or locate objects. Training-based methods face generalization challenges, while training-free methods exploit multimodal large language models (MLLMs) but often lack mechanisms to reconcile proposed actions with spatial evidence, task progress, and execution failures. We present HarnessVLN, a zero-shot, training-free framework whose Agent Harness coordinates perception, retrieval, grounding, navigation, recovery, and termination through a unified tool interface. The Harness validates planner proposals against spatial evidence, geometric feasibility, and subgoal consistency, incorporating structured tool feedback into subsequent decisions. Hierarchical event memory tracks task progress and execution history, while a persistent Spatiotemporal Graph maintains reusable spatial evidence and failure annotations for verification and recovery. A replaceable Navigation Executor converts validated targets into executable motions, allowing the same Harness protocol to support instruction-following and object-goal navigation. HarnessVLN achieves success rates of 60.8%, 53.9%, 76.0%, and 59.3% on R2R, RxR, HM3D-v2, and HM3D-OVON, respectively, surpassing prior training-free SOTA results. Humanoid deployment further demonstrates its applicability to both tasks in real-world environments. The project page is: https://harnessvln.netlify.app/.
The French Open Science Strategy (SSO) 2026–2033 aims to strengthen the transformations initiated with the two previous national plans. Covering eight years, it is structured around ambitions, challenges, strategic objectives, and four-year action plans. The Strategy seeks to achieve: enhanced research visibility and impact, increased academic sovereignty, more cumulative, collaborative, reproducible, responsible, and sustainable research, and research that fosters innovation and extends beyond academia. The SSO identifies several challenges: heterogeneity of practices across disciplines, resistance to data sharing, fragility of research infrastructures, quality concerns in scholarly publishing, limited recognition of research software, and insufficient communication of clinical trial results. The Strategy highlights key levers to address these challenges, including research evaluation and international coordination. Artificial intelligence (AI) presents both opportunities and issues to address. The SSO aims to develop open science while respecting research security, guided by the principle: "as open as possible, as closed as necessary." The four-year action plans are aligned with eight strategic objectives. Four objectives focus on practices, governance, tools, services, infrastructures, and France’s positioning in the European and international landscape. The remaining four objectives address publications, data, software, and clinical trial results. The 2026–2029 action plan breaks down these objectives into 30 actions and 75 measures. It includes new project, such as the creation of a national forge for research software, and it strengthens flagship initiatives like HAL, Recherche Data Gouv, and Software Heritage. The SSO is complemented by an assessment of France’s national open science policy. While significant progress has been made, substantial work remains. The Strategy outlines an ambitious path for the next eight years—one that envisions a more transparent, collaborative, and sovereign French research system, capable of addressing the research challenges of the 21st century.
