Current AI energy benchmarks measure consumption per model invocation. For agentic systems, where a single user goal may trigger multistep orchestration, tool calls, retries, and failure recovery, the invocation count is an implementation artifact rather than a task property, and per inference normalization misrepresen...
Deepak Panigrahy, Aakash Tyagi, Stephen Abkin et al.· 0 citations
This work investigates knowledge-driven, large language model (LLM)-based storytelling for older adults. We present a reflective storytelling agent integrating knowledge graphs, user modelling, and argumentation theory to guide person-tailored narratives. Argument mining supports inspection of argument structure, while...
Jayalakshmi Baskar, Vera C. Kaelin, Kaan Kilic et al.· 0 citations
Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning capabilities of large language models (LLMs), yet existing approaches predominantly rely on sparse outcome-level feedback, posing a substantial credit-assignment challenge in long-horizon agentic tasks. In this work, we study structured ag...
Huining Yuan, Zelai Xu, Huaijie Wang et al.· 0 citations
Circuit study aims to explain how language models (LMs) implement a task by localizing and interpreting a circuit: a computational subgraph responsible for a specific model's behavior. Existing methods for circuit localization are largely hypothesis-driven: they define a task through a dataset, assume that all examples...
Daking Rai, Mor Geva, Ziyu Yao· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
Aligning generative models with human preferences remains a central challenge. While RLHF is widely adopted, it relies on large-scale preference data. This limitation is particularly evident in evaluating instruction-guided image edits, even though humans can often learn such preferences from only a few examples. We pr...
Yuxuan Zhang, Cong Wei, Penghui Du et al.· 0 citations
MemQ is introduced, which applies TD ($\lambda$) eligibility traces to memory Q-values, propagating credit backward through a provenance DAG that records which memories were retrieved when each new memory was created.
Large Language Models (LLMs) have demonstrated remarkable capabilities in orchestrating tools for reasoning tasks. However, existing methods rely on a step-wise paradigm that lacks a global perspective, which causes error accumulation over long horizons and restricts generalization to unseen tools. To overcome these li...
Tairan Huang, Siyu Shang, Qiang Chen et al.· 0 citations
For years, we have built LLM serving systems like any other critical infrastructure, with a single general-purpose stack hand-tuned over many engineer-years to support every model and workload. In this paper, we take the opposite bet and let coding agents automatically synthesize bespoke serving systems for different u...
Keisuke Kamahori, Shihang Li, Simon Peter et al.· 0 citations
This work introduces OracleProto, a reproducible framework for evaluating LLM forecasting in a controlled, temporally masked environment, and retains resolved forecasting tasks as reusable evaluation resources under explicit temporal boundaries, supporting continual accumulation of forecasting datasets and systematic e...
Yi-Ding Ma, Cheng-Yu Ruan, Kai-Bo Huang et al.· 0 citations
Computer Architecture research has traditionally required the exploration of vast, highly complex design spaces through manual human effort. Large Language Models (LLMs) offer an opportunity to accelerate this process by enabling fast and automated implementation and testing of different architectural mechanisms and th...
Vasilis Kypriotis, Alexander Blasberg, Alexandros Moiras et al.· 0 citations
In reinforcement learning with verifiable rewards (RLVR), challenging math questions whose sampled groups remain uniformly incorrect yield no informative update, letting easier questions dominate actor updates. This imbalance can contribute to distribution sharpening: low-$k$ accuracy improves while problem coverage co...
LLM-based agent judges are an emerging approach to evaluating conversational AI, yet a fundamental uncertainty remains: can we trust their assessments, and if so, how many are needed? Through 960 sessions with two model pairs across 15 tasks, we show that persona-based agent judges produce evaluations indistinguishable...