ELO Ratings Questions

ELO Ratings Questions

Author: Noah Gift September 18, 2025 Duration: 3:39

Key Argument

  • Thesis: Using ELO for AI agent evaluation = measuring noise
  • Problem: Wrong evaluators, wrong metrics, wrong assumptions
  • Solution: Quantitative assessment frameworks

The Comparison (00:00-02:00)

Chess ELO

  • FIDE arbiters: 120hr training
  • Binary outcome: win/loss
  • Test-retest: r=0.95
  • Cohen's κ=0.92

AI Agent ELO

  • Random users: Google engineer? CS student? 10-year-old?
  • Undefined dimensions: accuracy? style? speed?
  • Test-retest: r=0.31 (coin flip)
  • Cohen's κ=0.42

Cognitive Bias Cascade (02:00-03:30)

  • Anchoring: 34% rating variance in first 3 seconds
  • Confirmation: 78% selective attention to preferred features
  • Dunning-Kruger: d=1.24 effect size
  • Result: Circular preferences (A>B>C>A)

The Quantitative Alternative (03:30-05:00)

Objective Metrics

  • McCabe complexity ≤20
  • Test coverage ≥80%
  • Big O notation comparison
  • Self-admitted technical debt
  • Reliability: r=0.91 vs r=0.42
  • Effect size: d=2.18

Dream Scenario vs Reality (05:00-06:00)

Dream

  • World's best engineers
  • Annotated metrics
  • Standardized criteria

Reality

  • Random internet users
  • No expertise verification
  • Subjective preferences

Key Statistics

MetricChessAI Agents
Inter-rater reliabilityκ=0.92κ=0.42
Test-retestr=0.95r=0.31
Temporal drift±10 pts±150 pts
Hurst exponent0.890.31

Takeaways

  1. Stop: Using preference votes as quality metrics
  2. Start: Automated complexity analysis
  3. ROI: 4.7 months to break even

Citations Mentioned

  • Kapoor et al. (2025): "AI agents that matter" - κ=0.42 finding
  • Santos et al. (2022): Technical Debt Grading validation
  • Regan & Haworth (2011): Chess arbiter reliability κ=0.92
  • Chapman & Johnson (2002): 34% anchoring effect

Quotable Moments

"You can't rate chess with basketball fans"

"0.31 reliability? That's a coin flip with extra steps"

"Every preference vote is a data crime"

"The psychometrics are screaming"


Resources

  • Technical Debt Grading (TDG) Framework
  • PMAT (Pragmatic AI Labs MCP Agent Toolkit)
  • McCabe Complexity Calculator
  • Cohen's Kappa Calculator

🔥 Hot Course Offers:

🚀 Level Up Your Career:

Learn end-to-end ML engineering from industry veterans at PAIML.COM


Noah Gift guides you through a year-long journey with 52 Weeks of Cloud, a weekly exploration designed for anyone building, managing, or simply curious about modern cloud infrastructure. Each episode digs into a specific technical topic, moving beyond surface-level explanations to offer practical insights you can apply. You’ll hear detailed discussions on the platforms that power the industry-like AWS, Azure, and Google Cloud-and how to navigate multi-cloud strategies effectively. The conversation regularly delves into the orchestration of these systems with Kubernetes and the specialized world of machine learning operations, or MLOps, including the integration and implications of large language models. This isn't just theory; it's a focused look at the tools and methodologies shaping how software is deployed and scaled today. By committing to this podcast, you're essentially getting a structured, expert-led curriculum that breaks down complex subjects into manageable weekly segments, all aimed at building a comprehensive and practical understanding of the cloud ecosystem.
Author: Language: English Episodes: 225

52 Weeks of Cloud
Podcast Episodes
Pattern Matching Systems like AI Coding: Powerful But Dumb [not-audio_url] [/not-audio_url]

Duration: 7:01
Pattern Matching Systems: Powerful But DumbCore Concept: Pattern Recognition Without UnderstandingMathematical foundation: All systems operate through vector space mathematicsK-means clustering, vector databases, and AI…
Comparing k-means to vector databases [not-audio_url] [/not-audio_url]

Duration: 8:10
K-means & Vector Databases: The Core ConnectionFundamental SimilaritySame mathematical foundation – both measure distances between points in spaceK-means groups points based on closenessVector DBs find points closest to…
K-means basic intuition [not-audio_url] [/not-audio_url]

Duration: 6:40
Finding Hidden Groups with K-means ClusteringWhat is Unsupervised Learning?Imagine you're given a big box of different toys, but they're all mixed up. Without anyone telling you how to sort them, you might naturally put…
Greedy Random Start Algorithms: From TSP to Daily Life [not-audio_url] [/not-audio_url]

Duration: 16:20
Greedy Random Start Algorithms: From TSP to Daily LifeKey Algorithm ConceptsComputational Complexity ClassificationsConstant Time O(1): Runtime independent of input size (hash table lookups)"The holy grail of algorithms"…
Hidden Features of Rust Cargo [not-audio_url] [/not-audio_url]

Duration: 8:52
Hidden Features of Cargo: Podcast Episode NotesCustom Profiles & Build OptimizationCustom Compilation Profiles: Create targeted build configurations beyond dev/release[profile.quick-debug] opt-level = 1 # Some optimizati…
Using At With Linux [not-audio_url] [/not-audio_url]

Duration: 4:53
Temporal Execution Framework: Unix AT Utility for AWS Resource OrchestrationCore MechanismsUnix at Utility ArchitectureKernel-level task scheduler implementing non-interactive execution semanticsPersistence layer: /var/s…
Assembly Language & WebAssembly: Technical Analysis [not-audio_url] [/not-audio_url]

Duration: 5:52
Assembly Language & WebAssembly: Evolutionary ParadigmsEpisode NotesI. Assembly Language: Foundational FrameworkOntological DefinitionLow-level symbolic representation of machine code instructionsMinimalist abstraction l…
Strace [not-audio_url] [/not-audio_url]

Duration: 7:23
STRACE: System Call Tracing Utility — Advanced Diagnostic AnalysisI. Introduction & Empirical Case StudyCase Study: Weta Digital Performance OptimizationDiagnostic investigation of Python execution latency (~60s initiali…
Free Membership to Platform for Federal Workers in Transition [not-audio_url] [/not-audio_url]

Duration: 3:53
Episode Notes: My Support Initiative for Federal Workers in TransitionEpisode OverviewIn this episode, I announce a special initiative from Pragmatic AI Labs to support federal workers who are currently in career transit…
Ethical Issues Vector Databases [not-audio_url] [/not-audio_url]

Duration: 9:02
Dark Patterns in Recommendation Systems: Beyond Technical Capabilities1. Engagement Optimization PathologyMetric-Reality Misalignment: Recommendation engines optimize for engagement metrics (time-on-site, clicks, shares)…