OPHIS Research Workflow

OPHIS is [[LiuZiming|Liu Ziming]]’s proposed structure for making research legible enough to train AI systems on it. In 149. 亲历中美 New Labs 资本狂潮，和清华刘子鸣聊：AI for AI、机制可解释性和 Max Tegmark, the acronym stands for Observation, Problem, Hypothesis, Intervention, and Speed up.

The source’s key move is to treat research as a language. If papers and lab work can be marked up as observations, problems, hypotheses, interventions, and acceleration moves, then Auto Research can train on more than final papers. The structure is also meant to capture the missing reasoning between failed experiments, intuitive leaps, and next actions.

Key Claims

• Research papers hide much of the thinking that produced them, especially false starts and local judgment.
• OPHIS creates a schema for turning research work into training data.
• A structured research language may let models learn when to propose hypotheses, run interventions, or speed up an experiment.
• The workflow is a bridge between human Research Taste and automated research systems.
• Liu treats OPHIS as an early requirement for AI For AI, not a finished research autopilot.

Connections

• [[LiuZiming|Liu Ziming]] and [[YuanhuanIntelligence|Yuanhuan Intelligence]] — source person and company context.
• AI For AI, Auto Research, and Recursive Self-Improvement — research-automation branch.
• Physics Of AI and Mechanistic Interpretability — scientific structure the workflow is meant to support.
• Meta-Model Training Curve Prediction, Training Autopilot, and Vibe Training — later model and product layers.
• Research Taste, AI Verification, and Training Compute Allocation — judgment, validation, and compute-triage constraints.