Back to corpus
research notebacklog referencescore 22
Cognitive Twin — The Reversal
Workspace document requiring curation.
Full HTML reader
Read the full artifact
Extracted abstract or opening context
## Essence Instead of injecting context into generic models, fine-tune models that already know our patterns. Twin Alpha (Qwen3-235B MoE) trained on 38K conversation turns learns Mo's voice, decision patterns, and technical preferences. The reversal: from context-first to model-first architecture. Frontier models become escalation-only.
## Context SFT training submitted to Together AI on Qwen3-235B-A22B-Instruct. 34K train + 3.8K val conversations. DPO data ready for Phase 2 (793 pairs). MiniMax M2.5 running on Vast.ai H100 as interim inference. The twin swarm concept: Alpha (coding), Beta (review), Gamma (tests/docs).
## Tags fine-tuning, cognitive-twin, qwen, together-ai, sft, dpo, model-first, reversal
Promotion decision
What has to happen next
Keep in the searchable backlog until it intersects a live paper or system.
Why this is not always a full paper yet
Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.