pareto
Datasets
All datasets matching “pareto”pareto-rail-rollouts
Pareto Rail benchmark rollouts
Full agent transcripts for every published run of the Pareto Rail level-generation benchmark: models one-shot a browser rail-shooter level from a written theme, and visitors rank the results blind on the public site. The benchmark's methodology, per-run provenance manifests (rendered prompt, sealed commits, gate results, measured cost), and the level source produced by each of these transcripts live in the main repository:… See the full description on the dataset page: https://huggingface.co/datasets/paulbatum/pareto-rail-rollouts.pareto-aligned-financial-llmtulu-3-stage-0-to-50pareto-actor
Dataset Information
# Nodes
# Edges
# Features
7,600
33,544
931
Usage
from huggingface_hub import hf_hub_download
hf_hub_download(repo_id="SauravMaheshkar/pareto-actor", filename="processed/actor.bin", local_dir="./data/", repo_type="dataset")
dataset, _ = dgl.load_graphs("./data/processed/actor.bin")
Thank you @severo for helping me figure out the usage.
Pre-processed as per the official codebase of https://arxiv.org/abs/2210.02016
Citations… See the full description on the dataset page: https://huggingface.co/datasets/SauravMaheshkar/pareto-actor.tulu-3-stage-50-to-75daily-paper-2026-07-11-verify-gated-fanout-pareto
Verify-Gated Fan-Out: Measuring the Cost-Quality Pareto Frontier of Adversarial Verification in Tiered Multi-Agent Skill Workflows
TL;DR — Controlled empirical measurement of adversarial verification for multi-agent fan-out: the cheapest tier (Haiku, N=1, $3.46) Pareto-dominates all 27 configurations, and increasing skeptic breadth beyond N=1 can harm safety due to a Condorcet-style vote-dilution effect.
ThakiCloud AI Research · 2026-07-11 · 📝 Tech blog (KO)… See the full description on the dataset page: https://huggingface.co/datasets/thaki-AI/daily-paper-2026-07-11-verify-gated-fanout-pareto.