arXiv AI
Jun 6
Benchmark Everything Everywhere All at Once
arXiv:2606. 06462v1 Announce Type: new Abstract: Benchmarks are fundamental for evaluating and advancing LLMs and MLLMs by providing standardized and explicit measures of performance.
Can LLM Annotators Match Human Consistency? Benchmarking Human Agreement and Optimizing Multi-Agent with Contrastive Knowledge
Related stories
arXiv AI
Aug 7
Benchmarking the Benchmarks: Evaluating Benchmarks for Conversational Agents
arXiv:2608. 06329v1 Announce Type: cross Abstract: Task-oriented conversational agents are evaluated using curated or automatically generated benchmarks, yet benchmark quality is rarely assessed.
arXiv Machine Learning
Aug 24
Designing a Robust LLM-Based Evaluation System for Agentic AI in Drug Discovery Through Human Alignment
The paper introduces an LLM-as-a-Judge framework for evaluating the outputs of an agentic drug discovery assistant, ChatInvent, deployed at AstraZeneca. It defines four quality dimensions—Completeness, Relevancy, Structural Clarity, and Scope Adherence—alongside deterministic Tool Call Correctness checks, and validates the judge against five expert annotators. After optimizing the best-performing judge with few-shot demonstrations, alignment with human majority votes improves from 0.80 to 0.86, and the framework reveals that informal question phrasing does not degrade output quality.
arXiv Machine Learning
Jul 2
REALM: Reliable Expertise-Aware Language Model Fine-Tuning from Noisy Annotations
arXiv:2604. 17289v2 Announce Type: replace Abstract: Supervised fine-tuning of large language models relies on human-annotated data, yet annotation pipelines routinely involve multiple crowdworkers of heterogeneous expertise.
arXiv AI
Jun 3
Fixing FOLIO and MALLS: Verified Annotations and an LLM-assisted Framework to Focus Human Relabeling
arXiv:2606. 02837v1 Announce Type: cross Abstract: Accurate translation from Natural Language to First-Order Logic (NL-to-FOL) underpins neurosymbolic AI systems and Natural Language Inference (NLI), making the quality of NL-to-FOL benchmarks essential -- yet these datasets have never been rigorously audited.
arXiv Computation and Language
Oct 8
Goldsmith: Gold-Loss-Guided Definition Optimization with an Agentic Annotation Harness
Goldsmith is an agentic pipeline that transforms a small set of expert-annotated calibration examples into a reusable structured annotation definition treated as a trainable textual object. Candidate definitions are evaluated on the same gold examples using an executable structured loss, and a large language model editor revises high-loss failures only when loss decreases. The approach outperforms direct rewriting, OPRO, APE, and PromptBreeder in prompt-optimization tests and enhances downstream annotation across typed span, pair-level relation, and fixed-trigger event-argument tasks when combined with retrieval, score-based routing, and human review.
arXiv Computation and Language
Sep 2
Post-hoc Alignment of LLM-judges to Human Judgment Distribution
The paper introduces NAPHA, a lightweight post‑hoc alignment method that improves large language model (LLM) predictions of human judgment distributions (HJD) by matching LLM output distributions to HJD through entropy‑based class assignment and specialized alignment models. Experiments on five datasets show that while LLMs perform near human‑level on hard‑label tasks, they struggle with soft‑label predictions, and NAPHA consistently enhances soft‑label accuracy, especially on high‑entropy instances. The study also demonstrates that better entropy class prediction can further boost NAPHA’s effectiveness.
arXiv AI
Aug 3
M3MAD-Bench: Multi-Dimensional Evaluation of Multi-Agent Debate Across Domains and Modalities
arXiv:2601. 02854v2 Announce Type: replace Abstract: As an agent-level reasoning and coordination paradigm, Multi-Agent Debate (MAD) orchestrates multiple agents through structured debate to improve answer quality and support complex reasoning.
arXiv AI
Sep 28
Agentick: A Unified Benchmark for General Sequential Decision-Making Agents
Agentick is a unified benchmark for sequential decision‑making agents that evaluates RL, LLM, VLM, hybrid, and human agents on 37 procedurally generated tasks across six capability categories, four difficulty levels, and five observation modalities via a single Gymnasium‑compatible interface. It includes a Coding API, oracle reference policies, pre‑built SFT datasets, a composable agent harness, and a live leaderboard. An evaluation of 27 configurations and over 90,000 episodes shows no single approach dominates, with GPT‑5 mini leading overall, PPO excelling in planning and multi‑agent tasks, and the reasoning harness boosting LLM performance by 3‑10×, while ASCII observations outperform natural language.
arXiv AI
Sep 7
MABPD: Multi-Agent Bias Probing & Detection via Structured Argument Debate
MABPD (Multi‑Agent Bias Probing & Detection) is a training‑free pipeline that uses three specialized large language model agents to analyze news articles from complementary perspectives and resolve disagreements via a Structured Argument Debate (SAD) protocol. SAD imposes an asymmetric burden of proof—biased claims lacking grounded textual evidence receive zero weight—along with role‑weighted voting and post‑consensus verification, replacing task‑specific supervised decision boundaries. Ablation studies show that the debate module alone accounts for up to a 10.6‑point F1 gain, and on the BABE benchmark MABPD attains 83.4% macro F1, within 0.7 percentage points of the supervised state‑of‑the‑art, while achieving 75.0% zero‑shot accuracy on the SemEval 2019 HyperPartisan corpus.
arXiv AI
Jun 2
EvoPool: Evolutionary Programmatic Annotation for Label-Efficient Specialized Supervision
arXiv:2606. 01617v1 Announce Type: cross Abstract: Large language models excel at general tasks but underperform smaller supervised models in specialized, high-stakes domains where training labels are costly.
arXiv AI
Sep 2
WorldBench: Culturally Grounded Benchmark for Multilingual Agents
WorldBench is a new multilingual benchmark that tests large language model agents on culturally grounded everyday workflows, offering 1,600 tasks in seven languages and eight cultures. The benchmark evaluates agents through structured sandbox actions and introduces Constrained Task Success (CTS), a metric that assesses task completion, minimal modification, and other complementary aspects via deterministic and LLM-as-a-Judge evaluations. Experiments show that even leading models achieve only 49.2% CTS, revealing significant gaps in correctness and state preservation across languages and cultures.