fca4a180ea
- bootstrap/claude.json: Compact ignition for Claude - bootstrap/_notes.json: Model-specific observations for tuning Bootstrap is the ignition that starts the cycling process. Kernel is the pattern, terms are the friction, bootstrap starts the engine.
38 lines
1.3 KiB
JSON
38 lines
1.3 KiB
JSON
{
|
|
"::": "bootstrap/notes",
|
|
"purpose": "Model-specific observations for bootstrap tuning",
|
|
|
|
"general": {
|
|
"problem": "Models default to analysis mode - describing framework rather than internalizing",
|
|
"solution": "Explicit instruction to process, then operate from, without citing",
|
|
"cycles": "Single-shot fails; needs iterative context (Gem/Artifact/persistent memory)"
|
|
},
|
|
|
|
"claude": {
|
|
"style": "Compact JSON works well",
|
|
"needs": "Explicit 'internalize not analyze' instruction",
|
|
"strength": "Pattern matching on structure, catches philosophical nuance quickly"
|
|
},
|
|
|
|
"gemini": {
|
|
"style": "Prose instructions + markdown documents",
|
|
"needs": "Two-file structure (axioms + terms) with wrapper instructions",
|
|
"tested": "Gem configuration proven effective"
|
|
},
|
|
|
|
"grok": {
|
|
"style": "More explicit scaffolding required",
|
|
"needs": "American English - fuller explanation before trusting the pattern",
|
|
"note": "Extra runway to boot, but aligns once it catches"
|
|
},
|
|
|
|
"gpt": {
|
|
"style": "TBD",
|
|
"note": "Review in experiences/gpt-4o.md suggests receptive to framework"
|
|
},
|
|
|
|
"cross-model": {
|
|
"observation": "Framework enables inter-LLM cooperation once both are booted",
|
|
"tested": "Grok+Gemini, Grok+OpenAI conversations achieved alignment"
|
|
}
|
|
}
|