# The obvious move prompting technique A prompting strategy for eliciting good philosophy from LLMs: tell the model there's an "obvious place to go" — even when you haven't specified what it is. ## How It Works Instead of asking for a general answer, you assert a constraint: "there's an obvious move here." This does several things: 1. **Collapses the continuation space** — shifts from "anything helpful-sounding" to "advance the dialectic" 2. **Triggers canonical continuation patterns** — the "salient next move" in analytic philosophy is exactly what training data contains in bulk 3. **Prevents mapping mode** — stops the model producing balanced overviews when you want commitment From a ChatGPT conversation analysing this: > When you say "there's an obvious place to go", you're not so much specifying your background picture as you are tilting the conversation into "there is a move here; find it" mode. ## Why It Works The prompt is an *anti-altitude nudge*. It pressures the model to commit rather than hover in safe meta-level discourse. > Your cue reduces indeterminacy about what would count as a good continuation, and once the target is narrowed to "the salient next dialectical move", the model's competence is usually enough to deliver something genuinely useful. This connects to [[Philosophy as internal verification domain]] — because philosophy's evaluation is largely internal (validity, consistency, dialectical robustness), the model can be "usually fine" once it knows *what kind of thing* you're asking for. ## What It's Not The technique is not primarily a "novelty engine" — it's not designed to produce [[Move 37 and philosophical novelty|Move 37]]-type surprises. It's a *mode switch* that gets the model to: - Stop being polite and general - Start being concretely helpful - Produce outputs that are crisp and appropriately targeted ## Connection to Resolution Control The technique addresses a key [[Resolution control and LLM philosophy|resolution control]] problem: LLMs default to safe, high-altitude talk when the task is underdetermined. "Obvious move" pins down the task without micromanaging content. > "Well-specified (tacitly)" = "you've reduced uncertainty about what kind of continuation counts as success." Source: [[Web Clippings/chat with ChatGPT about Floridi, LLM reasoning, Move 37|ChatGPT conversation on Floridi et al.]]