datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
eu-ai-act-article-50-scoreboard
Article 50 historical public-evidence snapshot
This work was produced through an AI-assisted workflow directed by the author. Historical work used Anthropic assistance; the retrospective correction uses OpenAI GPT-6, with separate bounded Gemini advice. All three providers have products in the scored set.
Purpose: provide the corrected paper's version 1.1 bundle under v1_1. Start with its README and correction note. The paper and deposit and GitHub repository identify the same… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/eu-ai-act-article-50-scoreboard.model-dependency
The Model Is a Dependency
This work was produced through an AI-assisted workflow directed and reviewed
by the author.
NM AI Research · ORCID 0009-0003-4213-7769 · Version 1.1 · CC BY 4.0
Purpose: provide the corrected working paper, stored outputs, calculation scripts and entropy explorer for the numeric-extraction study.
Read the corrected paper and correction record. The series DOI identifies the version family. The previous version 1.0 remains the historical paper and… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/model-dependency.ai-constraint-relay
AI Infrastructure: Commitments, Delivery and Access
Purpose: route readers to the current paper and its executable evidence bundle.
Version: v7.1. Concept DOI.
This work was produced through an AI-assisted workflow directed by the author.
The paper examines selected disclosed commitments, delivery stages and access obligations across the US, EU and UK. It uses dated records assembled through 14 September 2026, with later Irish-process clarifications. It does not establish an… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/ai-constraint-relay.fineweb-edu-threshold
What actually admits a document to FineWeb-Edu
Three operational details in a published quality filter, measured across 84 million
documents and 94 Common Crawl snapshots.
AI disclosure: the research is the author's; this text was drafted with AI assistance and reviewed by the author. The model, and the conflict it creates, are named in the Conflict of interest section below.
Paper (DOI): 10.5281/zenodo.21740081 ·
Code: github.com/NMAIResearch/fineweb-edu-threshold ·
CC BY 4.0… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/fineweb-edu-threshold.ceo-pay-scorecard
CEO Pay-vs-Delivery Scorecard: S&P 500 v2.2
A frozen, reproducible dataset for a descriptive audit of granted compensation, Compensation Actually Paid and shareholder return across a dated S&P 500 universe.
AI disclosure: the research is the author's; this text was drafted with AI assistance and reviewed by the author. The model, and the conflict it creates, are named in the Conflict of interest section below.
Author: NM AI Research, ORCID 0009-0003-4213-7769
Canonical concept… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/ceo-pay-scorecard.announced-vs-deliverable-power
Announced vs. Deliverable AI Power Demand
How much announced generation and data-centre load actually get delivered. A reproducible deflation calculator and cross-region referee combining PJM interconnection-queue completion data with PJM and ERCOT load-side operator screens. Published with the data and standard-library reproduction and verification scripts.
AI disclosure: the research is the author's; this text was drafted with AI assistance (Anthropic Opus 4.8-5.0 and Google… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/announced-vs-deliverable-power.ai-private-to-public-ledger
The AI Private-to-Public Ledger
A pre-registered benchmark for venture marks and liquidity realisation: scoring private AI valuations against primary SEC filings, IPO pricing, and post-listing lock-up mechanics.
AI disclosure: the research is the author's; this text was drafted with AI assistance (Anthropic Opus 4.8-5.0 and Google Gemini 3.7) and reviewed by the author. The models, and the conflicts they create, are named in the Conflict of interest and scope section.
Author:… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/ai-private-to-public-ledger.ai-water-tracker
AI Data-Centre Water Tracker
AI disclosure: the research is the author's; this text was drafted with AI assistance
and reviewed by the author. The models, and the conflicts they create, are named
in the Conflict of interest and scope section.
An open, reproducible referee for the water burden of AI data centres. It rebuilds the Water Consumption Impact index for ten sites from public data, self-checks each value against its source, and adds the two channels the original… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/ai-water-tracker.cost-watch
AI Cost Watch
Reproducible, primary-source analysis of the AI industry: whether the buildout's unit economics are holding. A recurring, dated note, where each issue states a falsifiable test for its forward call. Every figure is published with its data and a script that regenerates it, so any number can be checked at source.
AI disclosure: the research is the author's; this text was drafted with AI assistance and reviewed by the author. The model, and the conflict it creates… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/cost-watch.forecast-scorecard
AI Energy-Demand Forecast Scorecard
A reproducible audit of how the field forecasts data-centre electricity demand: how the published forecasts disperse, how they get revised, and whether they are transparent enough to reproduce. Primary-sourced, published with the data and a script that regenerates every figure.
AI disclosure: the research is the author's; this text was drafted with AI assistance and reviewed by the author. The model, and the conflict it creates, are named in… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/forecast-scorecard.divergence-registry
Divergence Registry
A versioned protocol and deterministic checker suite for recording methodological decisions and
comparing research runs.
AI disclosure. OpenAI GPT-5.6 Sol assisted with implementation and release maintenance. Google
Gemini 3.7 Flash assisted with maintenance and trial orchestration. Anthropic Opus 5 performed the
independent review of v0.2-rc1. The author retained all research and release decisions. Models are
not authors.
Published release. v0.2-rc2 is… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/divergence-registry.bot-energy-bound
How Much Electricity Does Bot Web Traffic Actually Use?
A transparent order-of-magnitude bound on the electricity used by bot web traffic, and an account of why no single honest figure exists in public data. Two independent methods land on an order of roughly 10 TWh per year while disagreeing by about 5.3 times, and the work corrects a widely-cited figure downward by roughly 15 to 20 times. Every input is published with the parameter set and a script that regenerates the bound.… See the full description on the dataset page: https://huggingface.co/datasets/NMAIResearch/bot-energy-bound.