The paper reports that agent evaluations often show a tool‑call rate of zero even when the model emits valid calls, because the interface censors the trajectory before downstream components see it. Experiments on BFCL v4 and tau‑bench demonstrate that swapping the serving adapter can change the observed call rate from 0.00 to 0.96/0.19 or from 0 to 636 calls, indicating that the interface—not the model—causes the discrepancy. A 98‑line preflight check is released to detect such silent failures, highlighting that tool‑call rates depend on the model‑interface stack rather than the model alone.
By Wenbo Wang
SV-WAM is a surround‑view world‑action model that keeps all six camera views while enabling efficient inference by discarding the video branch at deployment. It uses future‑video prediction as dense training supervision and an action‑centered causal mask to prevent action tokens from attending to future‑video tokens during joint denoising. A differentiable drivable‑area compliance regularizer penalizes vehicle‑footprint corners near or crossing drivable boundaries, improving safety and boundary awareness. Experiments on NAVSIMv2 and nuScenes show state‑of‑the‑art planning performance with low latency and strong zero‑shot transfer.
By Jinyang Wang, Shiwei Li, Junjian Wang, Zhiqiang Deng, Jianbin Gao, Yihang Zhao, Liu Liu, Yongjia Zhao, Jinlong Chen, Huirui Xu, Yifeng Pan, Kangwei Liu, Fan Ren, Ji Tao, Minghao Yang
The paper introduces the DevRev NL2SQL benchmark, comprising 900 execution‑verified queries that feature nested‑type and link‑graph structures, along with the Semantic Depth Score (SDS) to assess analytical reasoning depth. It also presents a cost‑aware single‑generation agentic architecture that includes schema selection, metadata retrieval, and error‑repair components tailored to nested enterprise schemas. On the DevRev benchmark, the system achieves 91.7% answer correctness, outperforming the next‑best baseline by 54.6 percentage points, and remains competitive on the Spider 2.0 Snowflake dataset.
GPT‑6 Astra is a new OpenAI model rolling out today to a limited set of organizations and soon to all ChatGPT Plus, Pro, Business, Enterprise users, and via the OpenAI API and AWS. It is priced at $10/million input and $50/million output, matching Claude Fable 5/5.1, and outperforms Fable on most OpenAI self‑reported benchmarks, achieving 99.9% on the ARC‑AGI 3 benchmark with a custom Provider Adapter harness. Astra excels in security tasks—scoring 100% on ExploitBench, 42.4% on ExploitGym, and 99.2% on SRE‑Bench—and handles long context well, hitting 100% on OpenAI’s eight‑needle benchmark at 256K–512K tokens and 96.3% at 512K–1M tokens, though it remains behind Fable on the Intelligence Index and Meta’s Muse Spark 1.3.
Terminal-Universe is a framework that converts large collections of terminal-based agent trajectories into reusable, executable environments. By replaying recorded file operations and filling missing files with a completion agent, it reconstructs the original workspace and generates new tasks, scaling them in breadth (cross-workspace queries) and depth (multi-round interactions). The resulting 37.3k task-sufficient environments enable significant performance gains when fine‑tuning language models on terminal‑centric benchmarks.
The paper introduces a low‑cost, open experimental platform for end‑to‑end autonomous driving on miniature Ackermann vehicles. It combines a physical vehicle, a printed urban track, data collection tools, trajectory registration, and a Webots digital twin to link simulation and real‑world experiments. Using command‑conditioned behavior cloning, the authors demonstrate that a neural policy can follow lanes with a mean cross‑track error of 6.1 cm, and that synthetic data plus a sim‑to‑real image translator improves performance on all track routes.
The paper introduces "environment evolution," a method that incrementally raises the difficulty of interactive environments for terminal agents using off‑policy updates. It derives three evolution directions tied to a multi‑turn learning objective and implements them via a loop‑engineered multi‑agent harness. Experiments with models such as Hy4 preview, Claude Opus 5, GPT‑5.6 Sol, Qwen3.6‑27B, and Qwen3.6‑35B‑A3B demonstrate that this approach consistently generates harder environments and yields significant performance gains on Terminal‑Bench 2.1.
DRACO introduces a method for fine‑grained credit assignment in long‑horizon reinforcement learning tasks that lack verifiable rewards. It dynamically generates multi‑criteria rubrics during training, scores them once per trajectory, and redistributes the resulting judgment over the steps responsible for each rubric to produce differentiated per‑step advantages. Experiments on AppWorld and Tau‑Bench show that DRACO outperforms baseline models and other rubric‑based approaches, achieving significant performance gains without relying on verifiers.
The paper identifies a flaw in Group Relative Policy Optimization (GRPO) called the spurious advantage, where the advantage estimator can incorrectly reward rollouts that reach correct answers by guessing rather than reasoning. This issue appears in bounded-answer tasks, open-answer sets with bounded sub-cases, and search agents with many paths to the same answer. The authors propose SIGNBALANCE, a new magnitude estimator that removes composition, maintains verifier sign, applies a global scale, and uses stop‑gradient per‑class rescaling to restore zero‑mean balance, achieving comparable or better performance on math and search benchmarks.
FiMI Banking presents a controlled Indian retail‑banking environment designed to train conversational agents that can answer product questions, handle account requests, and comply with strict operational and regulatory constraints. The system is built from vetted banking documents, structured ground truth, synthetic customer backgrounds, and banking tools. Evaluation of two post‑training methods—preference optimization and reinforcement learning with verifiable rewards—shows that preference optimization significantly increases safe behavior (out‑of‑scope refusal from 52% to 80%), while reinforcement learning improves edge‑case and order‑sensitive task performance and reduces token usage.
The paper introduces CAPA, a Collaborative Agent Predictive Architecture designed to improve large language model (LLM) participation in online meetings. CAPA tracks meeting state with a Perceiver, predicts conversation flow, decides when and what to speak, and generates contributions in the participant’s style, all while being calibrated by judges. In experiments on 137 AMI meetings, CAPA cuts the LLM’s silence rate from 51.4% to 2.5%, doubles credited recovery, and maintains low hallucination.
The paper introduces set difference captioning for autonomous driving datasets, aiming to generate natural‑language descriptions of differences between two image subsets. It adapts a two‑stage approach to focus on object‑centric patches, enabling attribution of differences to specific objects or categories. A new benchmark, AD‑Diff Bench, is presented to evaluate this method, especially for sparse, real‑world differences, and the authors provide open‑weight models and code for reproducibility.
The article "How to Solve the Right Problem in the Age of Agentic AI" presents a practical framework aimed at reducing uncertainty before agents accelerate implementation. It offers guidance on identifying and addressing the most relevant problems in the context of increasingly autonomous AI systems.
By Mike Huls
The paper introduces Local Updates, Global Learning (LUGL), a framework that separates data collection from model training, allowing non‑incremental learners such as gradient‑boosted trees (LightGBM) to be used in reinforcement learning for games. LUGL alternates between a local phase—where self‑play generates tabular updates—and a global phase—where these updates train a function approximator before resetting the table. Experiments on both perfect‑information and imperfect‑information games show that LightGBM agents perform competitively or better than neural‑network baselines like DQN and DeepCFR.
SV-WAM is a surround‑view world‑action model that keeps all six camera views for autonomous driving while enabling efficient inference by discarding the video branch during deployment. It uses future‑video prediction as dense training supervision and introduces an action‑centered causal mask to prevent future‑video tokens from influencing action tokens during joint denoising. A differentiable drivable‑area compliance regularizer further improves safety by penalizing vehicle‑footprint corners that approach or cross drivable boundaries. Experiments on NAVSIMv2 and nuScenes show state‑of‑the‑art planning performance with low latency and strong zero‑shot transfer.
KC-Bench is a dynamic interactive benchmark designed to evaluate how large language model agents reconcile user instructions, internal knowledge, and real‑time environmental observations. It consists of 238 manually curated multi‑turn tasks that test world‑knowledge conflicts, input inconsistencies, and multi‑source temporal conflicts, using a user simulator, stateful tools, deterministic environment assertions, an open‑source natural‑language evaluator, and human trajectory verification. Evaluation of nine models—including DeepSeek‑V4‑Flash, GLM‑5.2, and MiniMax‑M3—reveals significant cross‑domain variation, with no model reliably handling factual correction, identity consistency, and temporal conflict resolution across all settings, and shows that missed conflicts can propagate to tool calls or synthetic protected‑data flows.
Scalable Context Orchestration for Serving LLMs Over Voice presents llmovoice, a middleware that explicitly models voice context—including speaking rate, background noise, and packet loss—to guide large language model responses. By constructing a bounded voice context at each turn, llmovoice improves alignment with user preferences and reduces errors, achieving a 52.4% drop in speaking‑rate alignment error and a 0.9% false‑interruption rate under packet loss. In addition, it cuts model usage costs dramatically, lowering per‑turn cost by up to 24.9× while maintaining 98.7% of baseline answer quality in long sessions.
The paper reviews the evolution of AI recruitment systems from simple profile matching to complex, multi‑stage workflows that retrieve evidence, compare candidates, and execute actions. It organizes 40 representative works, highlighting transitions from similarity to reciprocal suitability, from single models to compound workflows, and from offline prediction to evidence‑aligned evaluation. The authors identify persistent gaps such as confounded behavioral labels, limited data validity, hidden pipeline failures, and a lack of privacy assessment, and propose a staged mapping for defensible evaluation and an agenda for evidence‑grounded, auditable systems.
Air-Ground Collaborative Vision-and-Language Navigation (AGC-VLN) pairs an unmanned aerial vehicle (UAV) with a global bird’s‑eye view and an unmanned ground vehicle (UGV) with a local first‑person view, creating a shared bird’s‑eye map that enables collaboration. The training‑free baseline decomposes navigation into VLM‑based semantic reasoning and deterministic geometric execution, allowing the UAV to render the UGV’s pose and target markers while the UGV plans a road‑following path using the shared map. In CARLA‑Air’s Town10HD scene, AGC‑VLN achieves a 77.0% joint success rate, a 27.0% improvement over the weaker individual agent and surpasses the strongest single‑agent baseline by 24.0 points.
The Civilization Framework proposes a new way to structure communication between AI agents by treating the civilization—comprising a human sovereign, a persistent ledger, and interchangeable agents—as the addressable party rather than individual agents. It introduces the Embassy Protocol, an asynchronous, carrier‑agnostic overlay that routes messages to a ledger endpoint where any online agent can process them, with commitment state on ledgers serving as the true record of interaction. The paper also identifies a temporal‑weight effect in AI‑to‑AI communication, demonstrates its impact in a preregistered experiment, and discusses mitigation strategies such as instruction‑level provenance labeling and sealed‑answer accuracy equivalence.
whyItMatters":"The framework offers a novel architecture that could reduce context loss and authority bias in multi‑agent AI systems, potentially improving reliability and accountability in AI‑driven interactions."