The Morning
From the arXiv
GIST-CMTF: Goal-State Inference for Causal Minimal Tool Filtering in LLM Agents
his paper introduces GIST-CMTF, a method to improve LLM agents' tool selection by first inferring the user's intended goal. It addresses the issue of "wrong-goal execution" by predicting potential goals and then applying causal minimal tool filtering (CMTF) or prompting for clarification if ambiguity exists. This approach significantly enhances task success rates for LLM agents by ensuring they pursue the correct objective.
OpenClaw-Skill: Collective Skill Tree Search for Agentic Large Language Models
This paper introduces Collective Skill Tree Search (CSTS), a novel framework for automatically constructing reusable skills for LLM agents. CSTS uses an iterative process of generating diverse skill candidates from multiple models and then having those models …
Skill-to-LoRA: From Using Skills to Learning Behaviors for Token-Efficient LLM Agents
This paper introduces Skill-to-LoRA (S2L), a novel method for representing and utilizing agent skills in Large Language Models (LLMs). Instead of relying on lengthy skill descriptions at runtime, S2L learns skill-specific LoRA adapters that capture the behavio…
TokenPilot: Cache-Efficient Context Management for LLM Agents
TokenPilot tackles the rising inference costs of LLM agents by efficiently managing their context. It employs a dual-granularity approach: **Ingestion-Aware Compaction** stabilizes prompt prefixes and filters environmental noise at the entry point, while **Lif…
Context-Aware RL for Agentic and Multimodal LLMs
This paper introduces ContextRL, a reinforcement learning method that enhances LLMs' ability to pinpoint crucial information within complex contexts. By training LLMs to select the correct supporting context from similar options, ContextRL encourages fine-grai…
Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures
This paper introduces **Contrastive-Difference CKA (CKA_Delta)**, a novel method to analyze how different language model architectures represent high-level concepts. CKA_Delta effe…
Speaking the Language of Science: Toward a General-Purpose Generative Foundation Model for the Natural Sciences
LOGOS unifies diverse scientific tasks by representing scientific objects and their spatial interactions as discrete tokens within a shared autoregressive framework. This "scientif…
Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models
This paper introduces "Adaptive and Explicit Safe" (AES), a method to leverage Large Reasoning Models' (LRMs) inherent ability to detect safety risks. AES first uses Supervised Fin…
Bayesian Inference and Decision Audits for Public Archives of Frontier AI Evaluations
This paper proposes a Bayesian inference framework to analyze public AI evaluation archives, recognizing they are incomplete time series rather than definitive leaderboards. The co…
Compositional Reasoning Depth Predicts Clinical AI Failure: Empirical Evidence Consistent with Transformer Compositionality Limits in Electronic Health Record Question Answering
This paper proposes "hop count," a measure of reasoning steps needed to answer clinical questions from EHRs, as a predictor of AI failure. The core method involves annotating quest…
The Town Square
This tech news story describes a personal homelab setup designed as an AI development platform, detailing the hardware and software used to run AI models locally.
Workshops
TeslaMate is a self-hosted data logger that collects and visualizes your Tesla's driving data, offering insights into your vehicle's usage and performance.
Agent-Reach provides a unified CLI to grant AI agents internet browsing capabilities across platforms like Twitter, Reddit, and YouTube, enabling them to read and search content without API fees.
