Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization

• URL: http://arxiv.org/abs/2603.18388v1
• Date: Thu, 19 Mar 2026 01:14:36 GMT
• Title: Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization
• Abstract summary: We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting.<n>A two-layer explore-exploit mechanism combines random restart and epsilon-greedy sampling further escapes local optima.
• Score: 1.8420836938108678
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Automatic prompt optimization (APO) has emerged as a powerful paradigm for improving LLM performance without manual prompt engineering. Reflective APO methods such as GEPA iteratively refine prompts by diagnosing failure cases, but the optimization process remains black-box and label-free, leading to uninterpretable trajectories and systematic failure. We identify and empirically demonstrate four limitations: on GSM8K with a defective seed, GEPA degrades accuracy from 23.81% to 13.50%. We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting, enabling semantically labeled hypotheses, parallel minibatch verification, and interpretable optimization trace. A two-layer explore-exploit mechanism combining random restart and epsilon-greedy sampling further escapes local optima. VISTA recovers accuracy to 87.57% on the same defective seed and consistently outperforms baselines across all conditions on GSM8K and AIME2025.

Related papers

• FLARE: Few-shot Learning-based Adaptive Reflective Engine [0.9932714016119233]
  We introduce FLARE (Few-shot Learning-based Adaptive Reflective Engine), a framework that leverages advanced reflective mechanisms to optimize instructions.<n>We evaluate our method across a diverse suite of benchmarks using the GPT-5 series of models.<n>Our results demonstrate that FLARE consistently outperforms GEPA, winning on every task-model pair.
  arXiv  Detail & Related papers  (2026-08-03T22:13:12Z)
• Perception, Verdict, and Evolution: Hindsight-Driven Self-Refining Forensics Agent for AI-Generated Image Detection [80.15317858033534]
  ForeAgent is an agentic forensics framework for AI-generated image detection with iterative self-evolution.<n>ForeAgent achieves state-of-the-art performance on the Chameleon benchmark, reaching 82.18% accuracy.<n>ForeAgent produces more consistent and causally grounded reasoning compared to GPT-5 and GPT-5-mini.
  arXiv  Detail & Related papers  (2026-06-25T02:59:33Z)
• ContraPrompt: Contrastive Prompt Optimization via Dyadic Reasoning Trace Analysis [0.6372261626436676]
  We introduce ContraPrompt, built on the observation that when a model fails but succeeds on a retry with feedback, the difference constitutes an optimization signal.<n>Unlike prior contrastive methods, we compare complete intermediate reasoning processes.<n>ContraPrompt beats GEPA on 11, ties on 41, and loses on 1 at equal budget.
  arXiv  Detail & Related papers  (2026-04-20T08:17:15Z)
• Synergistic Directed Execution and LLM-Driven Analysis for Zero-Day AI-Generated Malware Detection [0.12891210250935145]
  Weaponization of LLMs for automated malware generation poses an existential threat to conventional detection paradigms.<n>This paper introduces a novel hybrid analysis framework that combines emphconcolic execution with emphdeep-learning-based vulnerability classification.<n>A benchmark comprising 2,500 LLM-synthesized samples achieves 98.7% accuracy on conventional malware and 97.5% accuracy on AI-generated threats.
  arXiv  Detail & Related papers  (2026-03-10T00:25:41Z)
• PRISM: Pushing the Frontier of Deep Think via Process Reward Model-Guided Inference [6.568081870814357]
  PRISM is an inference algorithm that uses step-level verification to guide both population refinement and solution aggregation.<n>Across mathematics and science benchmarks, PRISM is competitive or outperforms existing DEEPTHINK methods.
  arXiv  Detail & Related papers  (2026-03-03T00:03:42Z)
• ODAR: Principled Adaptive Routing for LLM Reasoning via Active Inference [60.958331943869126]
  ODAR-Expert is an adaptive routing framework that optimize the accuracy-efficiency trade-off via principled resource allocation.<n>We show strong and consistent gains, including 98.2% accuracy on MATH and 54.8% on Humanity's Last Exam.
  arXiv  Detail & Related papers  (2026-02-27T05:22:01Z)
• Trajectory Guard -- A Lightweight, Sequence-Aware Model for Real-Time Anomaly Detection in Agentic AI [0.0]
  Trajectory Guard is a Siamese Recurrent Autoencoder with a hybrid loss function that jointly learns task-trajectory alignment via contrastive learning and sequential validity via reconstruction.<n>At 32 ms latency, our approach runs 17$-27times$ faster than LLM Judge baselines, enabling real-time safety verification in production deployments.
  arXiv  Detail & Related papers  (2026-01-02T00:27:11Z)
• Sem-DPO: Mitigating Semantic Inconsistency in Preference Optimization for Prompt Engineering [5.568436850698628]
  Sem-DPO is a variant of DPO that preserves semantic consistency yet retains its simplicity and efficiency.<n>We show that Sem-DPO keeps learned prompts within a provably bounded neighborhood of the original text.<n>On three standard text-to-image prompt-optimization benchmarks and two language models, Sem-DPO achieves 8-12% higher CLIP similarity and 5-9% higher human-preference scores (HPSv2.1, PickScore) than DPO.
  arXiv  Detail & Related papers  (2025-07-27T05:20:13Z)
• ESSA: Evolutionary Strategies for Scalable Alignment [8.418036456622158]
  We present ESSA, a gradient-free framework that aligns Large Language Models (LLMs) using only forward inference and black-box optimization.<n>ESSA improves the test accuracy of Qwen2.5-Math-7B by 12.6% on GSM8K and 14.8% on PRM800K, and raises the accuracy of LLaMA3.1-8B on IFEval by 22.5%.<n>In large-scale settings ESSA shows stronger scaling than gradient-based methods.
  arXiv  Detail & Related papers  (2025-07-06T16:23:07Z)
• Semantic-Preserving Adversarial Attacks on LLMs: An Adaptive Greedy Binary Search Approach [15.658579092368981]
  Large Language Models (LLMs) increasingly rely on automatic prompt engineering in graphical user interfaces (GUIs) to refine user inputs and enhance response accuracy.<n>We propose the Adaptive Greedy Binary Search (AGBS) method, which simulates common prompt optimization mechanisms while preserving semantic stability.
  arXiv  Detail & Related papers  (2025-05-26T15:41:06Z)
• Supervised Optimism Correction: Be Confident When LLMs Are Sure [91.7459076316849]
  We establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning.<n>We show that the widely used beam search method suffers from unacceptable over-optimism.<n>We propose Supervised Optimism Correction, which introduces a simple yet effective auxiliary loss for token-level $Q$-value estimations.
  arXiv  Detail & Related papers  (2025-04-10T07:50:03Z)
• Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization [78.82586283794886]
  $chi2$-Preference Optimization ($chi$PO) is an efficient offline alignment algorithm provably robust to overoptimization.<n>$chi$PO implements the principle of pessimism in the face of uncertainty via regularization.<n>$chi$PO's simplicity and strong guarantees make it the first practical and general-purpose offline alignment algorithm provably robust to overoptimization.
  arXiv  Detail & Related papers  (2024-07-18T11:08:40Z)
• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
  We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences.<n>To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model.<n>Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
  arXiv  Detail & Related papers  (2024-05-26T05:38:50Z)
• Self-Play Preference Optimization for Language Model Alignment [75.83359213697854]
  Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences. We propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game. Our approach, dubbed Self-Play Preference Optimization (SPPO), utilizes iterative policy updates to provably approximate the Nash equilibrium.
  arXiv  Detail & Related papers  (2024-05-01T17:59:20Z)
• SADet: Learning An Efficient and Accurate Pedestrian Detector [68.66857832440897]
  This paper proposes a series of systematic optimization strategies for the detection pipeline of one-stage detector. It forms a single shot anchor-based detector (SADet) for efficient and accurate pedestrian detection. Though structurally simple, it presents state-of-the-art result and real-time speed of $20$ FPS for VGA-resolution images.
  arXiv  Detail & Related papers  (2020-07-26T12:32:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.