Residual entropy beats prompt length
Errors should track task-relevant ambiguity more closely than the number of tokens written.
Research program
A science of task-relevant communication.
Machine Pidgin treats human–AI specification as a semantic communication problem: transmit the smallest abstraction that preserves the distinctions a task actually needs.
Task-Optimal Pidgin
Information-Theoretic Specification Across Human–AI Expressiveness Gaps, by Constantine Goltsev. Draft preprint.
The paper models latent intent, messages, context, machine action, task distortion, and human authoring cost. It derives limits on intent recovery, a minimal task-sufficient abstraction, a task-semantic rate–distortion converse, and a value-of-information rule for clarification.
The empirical program
We will compare free-form prompting, natural language with examples, SPEAR, and fully formal specification. Experiments should measure authoring time, authoring error, task regret, clarification count, and subjective burden.
Detail has a U-shaped cost
Too little detail raises ambiguity; too much raises human cost and creates false precision.
Abstraction improves transfer
Explicit PRESERVE and IGNORE fields should generalize better across models and domains.
Ask only when the answer matters
Value-of-information clarification should outperform both never asking and always asking.
Scientific stance
This is a theoretical position and an open empirical agenda—not a completed validation. Results, negative findings, protocol changes, and conflicts of interest should be public. AI may assist the work, but named humans remain responsible for claims and authorship.