Empirical AI Research & Audits
TheBestSI produces original empirical audits of frontier model benchmarks, publishes rigorous software evaluation methodologies, and analyzes compute scaling curves and expert timeline forecasts.
AI Benchmarks Audit
Deconstructing MMLU, HumanEval, and GPQA. Why automated benchmarks frequently fail to predict real-world engineering and reasoning performance.
Evaluation Methodology
Our 5-dimensional framework for testing AI software: task fidelity, contextual window limits, data governance, ergonomics, and cost predictability.
Superintelligence Timelines
Analyzing compute expansion, transformer scaling limits, power infrastructure bottlenecks, and expert consensus variance on AGI arrival.
Research articles
- research
Trump Wants to Rename AI “Super Intelligence.” What Actually Changes?
Executive Order 14434 orders federal agencies to scrub 'artificial intelligence' and adopt 'super intelligence.' Behind the rhetoric lies an administrative sleight of hand: changing federal branding while leaving the underlying legal definitions and technology exactly where they were.
Tal Slavin · Oct 10, 2026 · 9 min read
- research
Superintelligence and today's tools: keeping the terms apart
Why the word 'superintelligence' in a brand name, a headline, or a policy document does not make a current product superintelligent, and how published definitions actually differ.
Adam Green · Oct 10, 2026 · 10 min read
- research
Reading benchmark claims critically
A checklist of questions that determine whether a benchmark number tells you anything about the tool you plan to use, with worked examples from published research.
Adam Green · Oct 10, 2026 · 9 min read
- research
Benchmark contamination explained
What it is, how it happens, how researchers have tried to measure and avoid it, and why it can make a public benchmark score misleading.
Adam Green · Oct 10, 2026 · 9 min read
- research
What is an agentic workflow?
A plain-language definition, how agentic workflows differ from chatbots and from fully autonomous agents, the common ways they fail, and the controls that matter.
Adam Green · Oct 10, 2026 · 10 min read