docs(article): add Fable 5 to the prior-mapping matrix#591
Conversation
Probed Anthropic's newest model (claude-fable-5) clean-room on its release day with the five mapping prompts. It holds the richest Use-Case 2.0 prior of the four models — recalls the slices and the six principles by name — yet still hedges on Use-Case 3.0, strengthening the finding that a dense 2.0 prior does not buy a 3.0 one. Updates the matrix, findings 3 and 4, the method note, and the changelog. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Caution Review failedPull request was closed or merged during review No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
WalkthroughDieser Pull Request erweitert die Dokumentation um das Modell Fable 5 in der Experiment-Analyse „How Far the Prior Actually Reaches": Die Ergebnistabelle und Befunde werden um neue Spalten und Aussagen erweitert, die Setup-Anleitung wird präzisiert, und ein Changelog-Eintrag dokumentiert die Artikelaktualisierung. ChangesExperiment-Dokumentation mit Fable 5
Estimated code review effort🎯 2 (Simple) | ⏱️ ~8 minutes Possibly related PRs
Suggested reviewers
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
What
Adds Fable 5 (Anthropic's newest model,
claude-fable-5) as a fourth column in the prior-mapping matrix of the An Anchor Delivers Only as Far as the Prior Reaches article.Probed clean-room on its release day with the same five mapping prompts already used for Haiku 4.5 / Sonnet 4.6 / Opus 4.8 (
claude -p --strict-mcp-config --setting-sources ""from a neutral dir).Why it matters
Fable 5 holds the richest Use-Case 2.0 prior of all four models — yet still hedges on 3.0. That sharpens finding 4: a dense 2.0 prior does not buy a 3.0 one; the two sit at different densities in the training data, not at different model strengths. The newest, strongest model does not close the gap.
Changes
docs/training-data-vs-practice.adoc— Fable 5 column in the matrix; findings 3 and 4 updated; method note made precise (A–E experiment on Haiku + Opus, prior-mapping probes on all four)docs/changelog.adoc—2026-06-09"Updated article" entryThe A–E anchor experiment was not re-run on Fable 5; only the prior-mapping probes were, and the text says so.
🤖 Generated with Claude Code
Summary by CodeRabbit
Release Notes