AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Inference-Aware Prompt Optimization for Aligning Black-Box Large Language Models

Saaduddin Mahmud, Mason Nakamura, Kyle Hollins Wray, Shlomo Zilberstein

AAAI 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
jointly optimize prompt and inference scale (IAPO) instead of inference-agnostic prompt optimizationblack-box LLM prompt optimization · six environmentsaverage cost-adjusted reward: up to 25% over disjoint, up to 50% over prompt-onlyImproved
restrict Top-K screening to K=1 (near-decoupled prompt selection) (ablation)PSST · Summarizationaverage cost-adjusted rewardHurt
restrict Top-K screening to K=1 (ablation)PSST · CommonsenseQAaverage cost-adjusted rewardNo clear effect
add Top-K screening to PSST under small training budget (ablation)PSST · real-data tasks (MATH, CommonsenseQA, Helpful-Harmless, Summarization)average cost-adjusted rewardMixed
use PSST structure-aware arm elimination instead of UCB, softmax, epsilon-greedy, uniformIAPO policy learning · six environmentsaverage cost-adjusted rewardImproved
randomly assign N per query rather than optimizing N per context (ablation)TRIPLE · six environmentsaverage cost-adjusted rewardHurt
choose prompt by single-shot accuracy rather than under majority votingprompt selection · MATH, MVaccuracy at higher budgetHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy