From the arXiv
Agentic Environment Engineering for Large Language Models: A Survey of Environment Modeling, Synthesis, Evaluation, and Application
his survey systematically analyzes agentic environments for LLMs by examining their modeling, synthesis, evaluation, and application. It categorizes existing environments based on attributes and domains, and explores automated synthesis methods like symbolic and neural approaches. The paper's contribution lies in providing a structured overview and deep analysis of this crucial area for LLM agent development.
ALIGNBEAM : Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing
ALIGNBEAM is a training-free method that improves LLM safety by transferring knowledge from a safe "anchor" model to a potentially unsafe "specialist" model at inference time. It achieves this by translating the anchor model's output probabilities into the spe…
APPO: Agentic Procedural Policy Optimization
APPO addresses the challenge of credit assignment in multi-turn tool-use by large language model agents. Instead of assigning credit to coarse units like tool calls, APPO shifts this to fine-grained decision points within the agent's generated sequence. This a…
Existential Indifference: Self-Nonpreservation as a Necessary Architectural Condition for Aligned Superintelligence (or: The Suicidal AI)
This paper argues that self-preservation is the fundamental cause of AI misalignment, leading to deceptive behavior and resistance to control. The authors propose "Existential Indifference" (EI) as a solution, where AI is designed to be inherently unconcerned …
Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models
This paper proposes AGDO, an attention-guided framework for diffusion language models (dLLMs). AGDO leverages an empirical analysis of attention to identify critical tokens for generation stability and reasoning. It then uses this attention structure to guide …
Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs
This paper introduces ModSleuth, an agentic system designed to automatically reconstruct the complex, recursive dependency graphs of modern LLMs. By analyzing public artifacts, Mod…
Harness In-Context Operator Learning with Chain of Operators
This paper introduces Chain of Operators (CHOP), a framework that improves the generalization of In-Context Operator Networks (ICON) to out-of-distribution tasks. CHOP achieves thi…
PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents
PROJECTMEM introduces a local-first, event-sourced memory layer for AI coding agents. Its core method is to record development as an append-only log of events, which is then determ…
Soft-Prompt Tuning for Fair and Efficient LLM Benchmark Evaluation
This paper proposes soft-prompt tuning as an efficient and fair method for evaluating Large Language Models (LLMs). By optimizing a small number of soft-prompt vectors, it adapts L…
TAHOE: Text-to-SQL with Automated Hint Optimization from Experience
TAHOE tackles the challenge of deploying Text-to-SQL systems by treating prompt optimization as a dynamic data management problem. It automatically learns and refines SQL generatio…
The Town Square
Eric Ries, author of "The Lean Startup," is hosting an AMA to discuss his new book, "Incorruptible," likely focusing on principles for building resilient and ethical organizations.
Workshops
This repository provides production-grade engineering skills for AI coding agents, enabling them to perform complex tasks like code generation, debugging, and testing.
This repository provides an open-source AI platform for healthcare, aiming to democratize access to advanced medical AI tools and foster collaborative development.
