arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.21397

Watts If? A Benchmark for Counterfactual Reasoning About the AI Data Center Buildout

Pankaj Chouhan·Rajarshi Das·Aditya Dubey·Manminder Tomar·Sreekar Reddy·Samta Sharma·Satyapragnya Kar·PR Panigrahi
View original source ↗Read source document ↗Export metadata ↓

Abstract

An AI data center is committed years before it draws a watt: a few discrete choices (grid or on-site gas, evaporative or dry cooling, which county to file in) fix its capital, water draw, and carbon path, and LLMs are now asked to help screen them with no test of whether they track what each commitment rules out. We introduce DCBench-440, with two arms reported separately: 200 generative questions over STRIPS worlds for five US deal archetypes, with search-verified gold answers and no tested shortcut above 12% on any question type, and 240 multiple-choice questions grounded in quoted site records, emission equations, and stated scenarios. Across open-weight deployments, accuracy spans 12.5-71.5% (generative) and 58.3-92.5% (multiple choice). One skill fails everywhere: asked to list every action available at a decision gate, no model exceeds 4.0% exact match. The strongest list 72-85% of the actions that would change the project's state but omit most permitted actions that change nothing; with those excused, the best exact match is still 24%. Code, data, and other materials are available upon request.

Explore related subjects

cs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Pankaj Chouhan, Rajarshi Das, Aditya Dubey, Manminder Tomar, Sreekar Reddy, Samta Sharma, Satyapragnya Kar, PR Panigrahi. 2026-10-15. Watts If? A Benchmark for Counterfactual Reasoning About the AI Data Center Buildout. https://arxiv.org/abs/2610.21397

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Atomic Proximal Policy Optimization for Electric Robo-Taxi Dispatch and Charger Allocation

Pioneering companies such as Waymo have deployed electric robo-taxi services in several U.S. cities. Operating such fleets requires coordinating ride matching, vehicle repositioning, and battery charging in a stochastic environment while accounting for limited charging infrastructure. Understanding both how to design efficient, scalable control algorithms for large fleets and how charger deployment shapes system performance is therefore critical for robo-taxi operators. We model the operations of an electric robo-taxi system as a discrete-time Markov decision process with a long-run average reward objective. To efficiently evaluate system performance under different configurations, we develop a scalable deep reinforcement learning algorithm, Atomic Proximal Policy Optimization (Atomic-PPO), which reduces the action space using atomic action decomposition. The resulting training procedure requires only modest computational resources, making it practical to evaluate a large number of system configurations. We calibrate a simulation environment using real-world NYC for-hire vehicle trip records and benchmark Atomic-PPO against heuristic policies and a fluid-based reward upper bound. The results show that Atomic-PPO produces near-optimal fleet dispatching policies and serves as a reliable tool for system evaluation. We conduct extensive numerical experiments to evaluate how charging infrastructure and fleet size affect system performance. Our findings highlight the importance of ridership-based charger placement, balanced investments in charger quantity and charging speed, and joint planning of charging infrastructure and fleet size.

cs.AI↗
arXiv2026

Generalising from Self-Produced Data: Model Training Beyond Human Constraints

Current large language models (LLMs) are constrained by human-derived training data and limited by a single level of abstraction that impedes definitive truth judgments. This paper introduces a novel framework in which AI models autonomously generate and validate new knowledge through direct interaction with their environment. Central to this approach is an unbounded, ungamable numeric reward - such as annexed disk space or follower count - that guides learning without requiring human benchmarks. AI agents iteratively generate strategies and executable code to maximize this metric, with successful outcomes forming the basis for self-retraining and incremental generalisation. To mitigate model collapse and the warm start problem, the framework emphasizes empirical validation over textual similarity and supports fine-tuning via GRPO. The system architecture employs modular agents for environment analysis, strategy generation, and code synthesis, enabling scalable experimentation. This work outlines a pathway toward self-improving AI systems capable of advancing beyond human-imposed constraints toward autonomous general intelligence.

cs.AI↗
arXiv2026

Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis

Data synthesis for training large reasoning models offers a scalable alternative to limited, human-curated datasets, enabling the creation of high-quality data. However, existing approaches face several challenges: (i) indiscriminate generation that ignores the solver's ability and yields low-value problems, or reliance on complex data pipelines to balance problem difficulty; and (ii) a lack of reasoning in problem generation, leading to shallow problem variants. In this paper, we develop a problem generator that reasons explicitly to plan problem directions before synthesis and adapts difficulty to the solver's ability. We recover problem-design CoT from human-authored multi-part questions, using later subquestions as targets to infer the design rationale. These traces initialize the generator, which is then optimized with a seed-conditioned reward based on solver answer consistency. The reward guides difficulty adjustment relative to each seed while favoring problems near the solver's competence boundary. Across 10 mathematical and general reasoning benchmarks, our method outperforms seed-only training by 2.69 and 3.39 percentage points after one and three rounds, respectively. The framework also extends to vision-language models. Our code is publicly available at https://github.com/WalkerWorldPeace/Reasoning-Synthesis.

cs.AI↗