AMV Prompt Framework
AMV prompt framework is Li Jigang / 李继刚’s prompting model in E45 孟岩对话李继刚：人何以自处. A is the starting point or role position, V is the destination or direction, and M is the mental shape or reasoning path from A to V.
The point of the framework is that role and goal are not enough. If the user only sets a starting role and an endpoint, the model will interpolate in its own way through the answer space. M constrains how the model should think, which frames it should use, and what shape the reasoning should take.
The episode treats AMV as both technical and artistic. It is a practical way to specify prompts, but the difficult part is choosing the right parameters, frames, and paths, much like investing formulas depend on judgment about growth, risk, and context.
Key Claims
- A specifies where the model should start, often through role, stance, or context.
- V specifies the direction, desired end state, or vector of the answer.
- M specifies the mental path, thinking shape, or reasoning constraints between start and destination.
- Without M, the model may fill the middle with plausible but undesired reasoning.
- The framework works best when paired with rich Context Engineering rather than thin roleplay.
Connections
- Prompt As Intent Transmission — broader prompting theory that AMV operationalizes.
- AI Communication Ability — user skill needed to express A, M, and V clearly.
- Context Engineering — supplies the context that makes A and M concrete.
- Human Judgment Under AI — the user still evaluates whether the chosen path produced a fitting answer.
- AI Skills — AMV-style procedures can be turned into reusable agent instructions.