The Morning
From the arXiv
Collaborative Human-Agent Protocol (CHAP)
his paper introduces the Collaborative Human-Agent Protocol (CHAP), a new standard for managing complex, multi-agent and human collaborations. CHAP defines a shared workspace to facilitate seamless interaction and coordination between humans and AI agents, particularly in operational roles involving critical decision-making. Its core contribution is to provide a structured framework for these interactions, moving beyond current ad-hoc methods and enabling more robust and traceable human-AI teamwork.
PRISM: Recovering Instruction Sets from Language Model Activations
PRISM recovers the set of active instructions guiding a language model's behavior by decoding its internal activations. Its core method involves training an activation-conditioned interpreter using GRPO to directly predict instruction sets, rewarding accurate …
SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research
This paper introduces SearchSwarm, a method for enabling "delegation intelligence" in LLM agents for long-horizon research tasks. The core method involves a main agent decomposing tasks and delegating subtasks to subagents, which return summarized results to c…
SecureClaw: Clawing Back Control of LLM Agents
SecureClaw introduces a dual-boundary architecture to secure LLM agents. It protects against unauthorized actions by requiring a PREVIEW-COMMIT protocol for state-changing writes, allowing only a trusted executor to commit actions. Sensitive data is protected …
Rethinking the Divergence Regularization in LLM RL
This paper proposes Divergence Regularized Policy Optimization (DRPO) to improve the stability of Reinforcement Learning for Large Language Models (LLMs). DRPO replaces the hard masking used in previous methods with a smooth, advantage-weighted quadratic regul…
Streaming Interventions: Can Video Large Language Models Correct Mistakes as They Occur?
This paper introduces Ego-MC-Bench, a benchmark designed to evaluate video LLMs' ability to proactively correct user mistakes in real-time during everyday tasks like cooking. The c…
IS-CoT: Breaking the Long-form Generation Collapse via Interleaved Structural Thinking
This paper addresses the "length collapse" issue in LLMs generating long-form content. Their core method, IS-CoT, introduces a dynamic "Plan-Write-Reflect" cycle directly within th…
PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models
This paper introduces PsychoSafe, a novel framework for LLM refusals that moves beyond simple non-compliance. PsychoSafe reframes refusals as structured, psychologically-informed s…
The Neutral Mask: How RLHF Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language Model
This paper investigates how Reinforcement Learning from Human Feedback (RLHF) aligns large language models. The authors demonstrate that RLHF doesn't fundamentally alter a model's …
AGENTSERVESIM: A Hardware-aware Simulator for Multi-Turn LLM Agent Serving
AGENTSERVESIM is a hardware-aware simulator designed to evaluate multi-turn LLM agent serving policies. Its core method involves simulating agent execution at a program level, acco…
The Town Square
Apple has unveiled a new AI architecture that integrates Google's Gemini models, aiming to enhance its on-device AI capabilities.
Workshops
TurboVec is a Rust-based vector index leveraging TurboQuant, offering efficient vector storage and retrieval with convenient Python bindings.
This repository provides a collection of reusable agent skills designed to integrate with Google products and technologies, enabling AI agents to perform actions and access information within the Google ecosystem.
