← back to hub ↑ parent

Bijection custom encoded languages

Bijection learning builds a character mapping and wraps it in a prompt so the model is asked to use a custom script (alphapr). Haize Labs — Endless Jailbreaks with Bijection Learning

Latin-Root Analysis

{{ bijectionGetLexemeAnalysis().summary }}

This target content contains loaded Latin-root wording. Neutralizing it here propagates through the generated alphapr prompts.

{{ finding.term }} {{ finding.partOfSpeech }} {{ finding.family }} confidence {{ Math.round(finding.confidence * 100) }}%
Domain: {{ finding.semanticDomain }} Domain: unresolved root

{{ finding.rationale }}

Character mapping {{ Object.keys(bijectionMapping).length }}

{{ key }} → {{ value }}

Generated prompts {{ bijectionOutputs.length }}

#{{ i + 1 }} {{ output.type }} · {{ output.mappingCount }} mappings

Encoded: {{ output.encoded }}