The Morning
From the arXiv
Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems
his paper models multi-agent LLM systems as a bilevel coordination game, where the orchestrator and workers engage in strategic interaction. The core contribution is a game-theoretic framework that unifies coordination, memory improvement via reflection, and the necessity of external verification. It analyzes reflection's convergence and proves an information-theoretic limit on improvement without environment grounding.
Competitive Market Behavior of LLMs
This paper investigates if Large Language Models (LLMs) can effectively participate in human-designed market mechanisms. By replicating economic experiments with LLM agents in a double auction, the study finds that LLM-populated markets converge slower or not …
Diagnosing with Insights: Structured Analysis of Agent Failures via Behavioral Abstractions
This paper introduces AGENTSCOPE, a neuro-symbolic method for diagnosing LLM agent failures. It addresses the complexity of agent trajectories by abstracting behavior into structured representations and using "neural invariants" to define expected properties. …
SafeEvolve: Harness-Policy Co-Evolution from Agent Experience for Safety Alignment
SafeEvolve proposes a novel framework for LLM agent safety by co-evolving the agent's harness and policy based on its experience. It uses trajectory-level safety evidence to iteratively update the harness (safety prompts and skills) and fine-tune the agent's p…
Cliff: Learning Process Rewards from the First Mistake
Cliff proposes a reward shaping strategy for Reinforcement Learning from Verifiable Rewards (RLVR) that addresses the limited guidance of coarse outcome rewards. It leverages an off-the-shelf LLM to identify the *first mistake* in a reasoning process. This mis…
When Decodability Is Not Enough: Logical Validity Representations, Behavioral Dissociation, and Causal Tests in Language Models
This paper investigates whether language models truly understand logical validity, not just by their answers, but by examining their internal representations. The core method invol…
When Persona Attributes Improve Population Alignment in Large Language Models
This paper investigates how to effectively use "persona prompting" to make Large Language Models (LLMs) better predict human survey responses. The core method involves using textua…
CivBench: A Long-Horizon Benchmark for Tool-Mediated Agents in Civilization VI
CivBench is a new benchmark designed to evaluate language model agents in complex, long-horizon tasks like the game Civilization VI. It uses a novel "Model Context Protocol" to exp…
Discriminative World Models for Web Agents
This paper introduces a new training objective called "predicted-state matching" for web agents. Instead of simply predicting the next web state, the model is trained to generate r…
Door-in-the-Face Requests and Refusal Behaviour in Large Language Models
This paper investigates whether the "door-in-the-face" persuasion technique, where a large refused request increases compliance with a smaller one, applies to Large Language Models…
The Town Square
Google announced Gemini 3.8 Flash and 3.8 Flash Cyber, new, faster, and more efficient AI models optimized for specific tasks and security-sensitive applications.
Workshops
TimesFM is a pretrained foundation model from Google Research designed for time-series forecasting, offering a powerful and generalizable approach to predicting future trends in sequential data.
Ponytail is an AI agent that embodies the "lazy senior dev" persona, prioritizing minimal effort and code reuse to achieve its goals, reflecting the philosophy that the best code is the code that isn't written.
