We present X-Talk, an open-source framework that champions a decoupled, modular design for LLM-driven speech-to-speech (S2S) systems. While the dominant trend favors end-to-end (E2E) modeling to optimize information flow, these "omni-models" often struggle to balance the competing objectives of complex speech tasks within a single network. X-Talk challenges this paradigm by demonstrating that a systematically optimized cascaded pipeline can achieve sub-second latency without sacrificing modular flexibility. Our framework seamlessly integrates specialized front-end components (e.g., VAD, speech enhancement) and diverse understanding models (e.g., ASR, emotion, and environmental sound analysis) with LLM capabilities like retrieval-augmented generation (RAG) and tool use. By revitalizing the cascaded approach, X-Talk highlights the underestimated potential of modular S2S systems and provides a robust foundation for future research and applications.
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.
In token-priced language model services, users can see the prompts they submit and the answers they receive, but may also pay for internal reasoning and candidate answers that are not shown in the final response. This paper develops a three-party service model to study how charging for this intermediate computation shapes providers' decisions and social welfare. Users purchase answers from a model firm that pays an execution cloud, with either supplier controlling the computation budget. The model separates user billing from upstream payments and identifies when metered pricing encourages excessive generation, when fixed service prices restrict valuable computation, and how responsibility for execution costs changes these choices. A LiveCodeBench study measures answer quality, latency, token use, and resource consumption across five model deployments, with numerical billing comparisons for the three covered by archived tariffs. At equal average charges on the reference workload, the calibrated comparisons show that billing structure changes both service participation and the computation budgets suppliers select. For accepted services with different budgets, the measured outcomes define the user-value thresholds at which their welfare ranking reverses. Metered pricing yields higher welfare when the additional inference it induces creates net social value, whereas fixed service pricing is preferable when the costs of that computation outweigh its benefits.
Chao Feng, Zihan Liu, Siddhant Gupta, Jan von der Assen
原文摘要
Hardware-in-the-Loop (HIL) testing is essential for automotive validation but suffers from fragmented and underutilized test artifacts. This paper presents HIL-GPT, an industry-deployed retrieval-augmented generation (RAG) system that integrates semantic retrieval with domain-adapted large language models to support test engineers in real-world HIL workflows. The system combines domain-specific embeddings to enable traceable retrieval of test cases and requirements under industrial latency and cost constraints. Through empirical evaluation, we show that compact, domain-adapted models can achieve a favorable trade-off among accuracy, latency, and cost compared to larger general-purpose models, challenging the assumption that larger models are always preferable in industrial NLP systems. An A/B user study further confirms that HIL-GPT improves perceived helpfulness, truthfulness, and satisfaction over general-purpose LLMs.
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track) · 2026 · 正式发表