Uses search-agent reading traces and tiered distractors to train LLMs for long-context, multi-hop reasoning, and introduces a rubric reward that supervises entity-level steps (applied only to correct finals). Improves evidence-grounded reasoning and resists reward hacking across 4B–30B models.
Proposes TrOPD, a method that restricts token-level on-policy distillation to regions where teacher supervision is reliable to stabilize training under teacher–student distribution mismatch. Adds outlier handling (clipping, masking, forward-KL) and off-policy guidance; shows consistent gains on math reasoning, code generation and general benchmarks.
Studies small trainable adapters (PEFT) used as persistent personal models on top of large foundation models, analyzing three scaling axes—Scale Up, Scale Down, Scale Out—and introducing MinT, an infrastructure for adapter identity, provenance, evaluation, and serving.
AI Agent Papers2026
Localizes harmful span-level errors inside long research-agent trajectories to show which trajectory segments make final answers unreliable. Provides a 1,000-instance TELBench of annotated spans and DRIFT, a claim-centric auditing method that improves span-level localization and first-error accuracy by up to 30 percentage points.
Analyzes how single-domain RL fine-tuning on LLMs induces cross-domain interference and shows this damage concentrates in a low-dimensional shared conflict subspace; proposes a local perturbation theory and short domain "refresh" procedures that selectively recover earlier domains with minimal collateral loss.
AI Agent Papers2026
A benchmark for evaluating web-browsing agents in Korean contexts, composed of 400 tasks (300 manually verified by native speakers). Includes a human-verified split and an adversarial synthetic split to probe failure modes; reveals large performance gaps for both frontier and Korean models.
AI Agent Papers2026
Enables agents to proactively discover multiple hidden problems in a user context and pair each with supporting evidence and concrete actions. Uses iterative discovery (batch rounds conditioned on prior finds) and reusable "thought templates" to expand coverage and ground claims.
AI Dataset2026
Provides ~1M synthetic Salvadoran‑Spanish personas (148k records, ~300M tokens) grounded in 2024 census distributions for demographics, occupations and locations; intended for training/evaluating localized LLMs and synthetic-data workflows. CC BY 4.0, adults only.
AI Dataset2026
Around 80K short audio clips paired with transcripts in JSON, organized for easy loading with the Hugging Face datasets ecosystem—designed for short-form speech tasks (ASR, TTS, fine-tuning) and quick prototyping with common Python data tools.
AI Agent Papers2026
Represents episodic memory as a Cue–Tag–Content graph and integrates LLM reasoning into active retrieval so agents iteratively reconstruct and prune evidence paths for long-horizon questions. Reports up to 23% gains on LoCoMo / LongMemEval while reducing token and runtime costs.
Benchmark for evaluating proactive LLM mediators in realistic, multi-domain conflict scenarios by constructing cases from real disputes, probing five socio-cognitive adaptation axes, and using a topic-localized evaluator that achieves 0.82 alignment with human experts.
AI Dataset2026
Provides 600,000 synthetic Vietnamese persona texts (100,000 records, 6 personas per record) aligned to Vietnam's 2024 census and surveys for training and evaluating NLP / text-generation models; includes 21 demographic and persona fields, CC BY 4.0, single train split.