# Scaffolding gradient as competence metric
Philosophical competence in LLMs can be measured by how much prompting/scaffolding is needed before the model reliably produces good work. This turns "LLM philosophical ability" into an empirical question about constraint engineering.
## The Research Question
> To what degree can current LLMs be prompted into producing "Good Philosophy," and what does the required "Scaffolding" reveal about their internalisation of the Space of Reasons?
## The Metric
We measure "Philosophical Competence" by the **Scaffolding Gradient**:
| Level | Description | Scaffolding Required |
|-------|-------------|---------------------|
| **Level 1** | Generic | Massive hand-holding (checklists, logic-flags) |
| **Level 2** | Dialectical | Responds to [[Resolution control and LLM philosophy|Resolution Control]] ("be more precise") |
| **Level 3** | Autonomous | Produces a [[Move 37 and philosophical novelty|Move 37]] with a single "[[The obvious move prompting technique|Obvious Move]]" nudge |
## Why This Matters
This framing:
- Acknowledges that AI's "Internalisation" is a **matter of degree**
- Provides a testable-looking thesis without pretending to solve the metaphysics of understanding
- Treats prompting as **constraint engineering** rather than content engineering
- Makes "LLM philosophical ability" about how much constraint engineering the user must supply
## The Goal
If the model can satisfy dialectical and logical constraints with less external scaffolding, it has more deeply internalised the norms of philosophical continuation.
A "better" philosophy-capable LLM would be one that requires less "Resolution Management" to stay precise.
Source: [[Web Clippings/chat with ChatGPT about Floridi, LLM reasoning, Move 37|ChatGPT conversation on Floridi et al.]]