The Morning
From the arXiv
A Zeroth-Order Paradigm for LLM Preference Alignment
his paper introduces Comparison-based Preference Optimization (ComPO), a novel zeroth-order method for aligning LLMs with human preferences. Instead of relying on direct optimization of differentiable losses, ComPO extracts directional information from preference pairs using comparison oracles. This approach offers computational efficiency and provides a convergence guarantee, with an online variant further enhancing its applicability.
CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents
CERA-MoA introduces a novel iterative reinforcement learning framework that co-evolves a dynamic router with continually learning LLM agents. Its core contribution is a predictive familiarity estimator that uses mid-layer hidden states to assess agent competen…
Cognitive Extensions for Dual-Process Language Agents: Memory and Self-Reflection in Interactive Environments
This paper introduces cognitive extensions, an Adaptive Memory Module (AMM) and a Self-Reflection Module (SRM), to enhance dual-process language agents in interactive environments. The AMM provides salience-gated memory, while the SRM offers execution-time val…
Higher-order pruning of experts in mixture-of-experts language models
This paper introduces HOPE (Higher-Order Pruning of Experts), a novel method for reducing the parameter count in Mixture-of-Experts (MoE) language models. Unlike prior methods that prune experts independently, HOPE considers the cooperative nature of expert us…
Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents
This paper introduces SynAgent, a framework for autonomous materials synthesis that uses multimodal Large Language Model (LLM) agents. Unlike traditional black-box optimizers, SynAgent's core method involves the LLM agents maintaining and evolving an explicit,…
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-Making
This paper investigates whether in-context learning (ICL) in LLM agents performing interdependent decision-making relies on genuine reasoning or statistical pattern extrapolation. …
STRETCH the Boundaries: A Unified Self-Taught Framework for Progressive LLM Evolution
STRETCH is a unified framework for progressive LLM self-improvement that addresses capability stagnation by dynamically adjusting challenge difficulty. Its core method, the Stretch…
Ask the Tool, Don't Guess: Agent Tool Calls Hold Their Progress, and the Serving System Should Read It
This paper addresses the inefficiency of agent systems waiting for tool execution by proposing that tools explicitly report their progress. Instead of guessing tool durations, the …
ASLEval: Measuring Privacy Exposure Displacement in LLM Agent Sessions
This paper introduces "privacy exposure displacement" to address limitations in evaluating LLM agent privacy. Their framework, ASLEval, measures the mismatch between local privacy …
Beyond Outcomes: Dual-View Relational Learning for Efficient Agent Benchmarking
This paper introduces DualViewEval, a novel method for compressing agent benchmarks. It addresses the high cost of evaluating agents by jointly learning from both final outcomes an…
The Town Square
The lead developer for PS5 Linux has resigned, citing a decline in open-source projects due to inexperienced users relying on LLMs they don't comprehend.
Workshops
This GitHub repository provides a coding agent skill designed for multi-phase security audits, generating independently verified and machine-readable findings.
Colibri is a C-based, zero-dependency inference engine that enables running large Mixture-of-Experts (MoE) models on personal hardware by streaming experts from disk.
