arXiv subjects

Adi Asija

Publications and source records attributed to Adi Asija.

4 recordsLinked to original sources

arXiv2025

Hell or High Water: Evaluating Agentic Recovery from External Failures

As language model agents are applied to real world problems of increasing complexity, they will be expected to formulate plans across large search spaces. If those plans fail for reasons beyond their control, how well do language agents search for alternative ways to achieve their goals? We devise a specialized agentic planning benchmark to study this question. Each planning problem is solved via combinations of function calls. The agent searches for relevant functions from a set of over four thousand possibilities, and observes environmental feedback in the form of function outputs or error messages. Our benchmark confronts the agent with external failures in its workflow, such as functions that suddenly become unavailable. At the same time, even with the introduction of these failures, we guarantee that the task remains solvable. Ideally, an agent's performance on the planning task should not be affected by the presence of external failures. Overall, we find that language agents struggle to formulate and execute backup plans in response to environment feedback. While state-of-the-art models are often able to identify the correct function to use in the right context, they struggle to adapt to feedback from the environment and often fail to pursue alternate courses of action, even when the search space is artificially restricted. We provide a systematic analysis of the failures of both open-source and commercial models, examining the effects of search space size, as well as the benefits of scaling model size in our setting. Our analysis identifies key challenges for current generative models as well as promising directions for future work.

cs.CL↗

arXiv2025

Learning To Communicate Over An Unknown Shared Network

As robots (edge-devices, agents) find uses in an increasing number of settings and edge-cloud resources become pervasive, wireless networks will often be shared by flows of data traffic that result from communication between agents and corresponding edge-cloud. In such settings, agent communicating with the edge-cloud is unaware of state of network resource, which evolves in response to not just agent's own communication at any given time but also to communication by other agents, which stays unknown to the agent. We address challenge of an agent learning a policy that allows it to decide whether or not to communicate with its cloud node, using limited feedback it obtains from its own attempts to communicate, to optimize its utility. The policy generalizes well to any number of other agents sharing the network and must not be trained for any particular network configuration. Our proposed policy is a DRL model Query Net (QNet) that we train using a proposed simulation-to-real framework. Our simulation model has just one parameter and is agnostic to specific configurations of any wireless network. It allows training an agent's policy over a wide range of outcomes that an agent's communication with its edge-cloud node may face when using a shared network, by suitably randomizing the simulation parameter. We propose a learning algorithm that addresses challenges observed in training QNet. We validate our simulation-to-real driven approach through experiments conducted on real wireless networks including WiFi and cellular. We compare QNet with other policies to demonstrate its efficacy. WiFi experiments involved as few as five agents, resulting in barely any contention for the network, to as many as fifty agents, resulting in severe contention. The cellular experiments spanned a broad range of network conditions, with baseline RTT ranging from a low of 0.07 second to a high of 0.83 second.

cs.MA↗

arXiv2025

Hidden in the Haystack: Smaller Needles are More Difficult for LLMs to Find

Large language models (LLMs) face significant challenges with needle-in-ahaystack tasks, where relevant information ("the needle") must be drawn from a large pool of irrelevant context ("the haystack"). Previous studies have highlighted positional bias and distractor quantity as critical factors affecting model performance, yet the influence of gold context size, the length of the answer-containing document, has received little attention. We present the first systematic study of gold context size in long-context question answering, spanning three diverse benchmarks (general knowledge, biomedical reasoning, and mathematical reasoning), eleven state-of-the-art LLMs (including recent reasoning models), and more than 150K controlled runs. Our experiments reveal that LLM performance drops sharply when the gold context is shorter, i.e., smaller gold contexts consistently degrade model performance and amplify positional sensitivity, posing a major challenge for agentic systems that must integrate scattered, fine-grained information of varying lengths. This effect persists under rigorous confounder analysis: even after controlling for gold context position, answer token repetition, gold-to-distractor ratio, distractor volume, and domain specificity, gold context size remains a decisive, independent predictor of success. Our work provides clear insights to guide the design of robust, context-aware LLM-driven systems.

cs.CL↗

arXiv2025

Tur[k]ingBench: A Challenge Benchmark for Web Agents

Can advanced multi-modal models effectively tackle complex web-based tasks? Such tasks are often found on crowdsourcing platforms, where crowdworkers engage in challenging micro-tasks within web-based environments. Building on this idea, we present TurkingBench, a benchmark consisting of tasks presented as web pages with textual instructions and multi-modal contexts. Unlike previous approaches that rely on artificially synthesized web pages, our benchmark uses natural HTML pages originally designed for crowdsourcing workers to perform various annotation tasks. Each task's HTML instructions are instantiated with different values derived from crowdsourcing tasks, creating diverse instances. This benchmark includes 32.2K instances spread across 158 tasks. To support the evaluation of TurkingBench, we have developed a framework that links chatbot responses to actions on web pages (e.g., modifying a text box, selecting a radio button). We assess the performance of cutting-edge private and open-source models, including language-only and vision-language models (such as GPT4 and InternVL), on this benchmark. Our results show that while these models outperform random chance, there is still significant room for improvement. We hope that this benchmark will drive progress in the evaluation and development of web-based agents.

cs.AI↗