Engineering
Safin-1: Safety from Within through Memory-Native State Evolution
Safin-1 is a foundation-model family that embeds safety as an intrinsic capability through Memory-Anchor Routing across Context History (MARCH). The…
Engineering
AgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling
AgentJudgeBench is a benchmark of 3,808 synthetically generated records spanning six DAG topologies and three difficulty tiers, built to systematically…
Engineering
DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory
DiagEvo is a self-evolution framework that derives its training curriculum entirely from information produced during self-play rather than external task…
Engineering
Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement
Harness-of-Harness (HoH) lets LLM-based coding agents autonomously develop software across iterative planning-coding-testing loops, balancing bug fixes against new capabilities and…
Engineering
EM^2Mem: Event-Centric Multimodal Memory for Large Language Models
EM^2Mem is an event-centric multimodal memory framework for long-video question answering that binds multimodal evidence to event anchors during memory…
Engineering
Control-Data Flow Separation: Stable Prompt Optimization in Multi-Agent LLMs
This paper introduces control-data flow separation, decoupling execution-critical protocols from task-relevant content in multi-agent LLM systems by representing routing, formatting,…
Engineering
InternReviewer & InternAdvocate: Objective Reward and Evaluation for Agentic Reinforcement Learning
InternReviewer and InternAdvocate are specialized agents for generating peer reviews and rebuttals, trained on a large-scale scholarly dataset with integrated…
Engineering
E-Commerce Bench: Evaluating LLM Agents on Long-Horizon Autonomous Business Operation
E-Commerce Bench is an open-source benchmark testing LLM agents on long-horizon business operations simulating a full year of e-commerce activity,…
Engineering
Adapting Without Gradients: Affine Statistics Transport and Certificate Analysis
This paper introduces CASTER, a gradient-free test-time adaptation method for frozen neural network models that stores source class statistics and…
Engineering
ReFlowSET: Representation-Aligned Latent Flow Matching for SAR-to-EO Image Translation
ReFlowSET is a conditional latent flow-matching framework for converting synthetic aperture radar (SAR) imagery into electro-optical (EO) imagery. Rather than…
Engineering
Learning Where Outcomes Change: Credit-Addressable Reasoning for Multimodal Geometry
This paper introduces credit-addressable reasoning for multimodal geometry tasks, aligning the semantic units used during inference with the learning mechanism…
Engineering
Token-Efficient Data Reasoning Agents via Adaptive Structuring of Unstructured Data
This paper introduces agentic data cracking, a method that dynamically structures unstructured data as a byproduct of agent reasoning: when…