APEX: New Framework Improves Prompt Optimization for Large Language Models Through Dynamic Data Selection
Researchers introduced APEX, a framework that optimizes prompts for large language models while dynamically selecting the most informative training data. The method categorizes data into Easy, Hard, and Mixed tiers and prioritizes Mixed tier examples where models show inconsistent performance. Under a fixed budget of 5,000 evaluations, APEX improved performance by 11.2% on Gemini 2.5 Flash and 6.8% on Gemma 3 27B, suggesting data-centric approaches are more efficient than traditional evolutionary algorithms.
APEX (Automatic Prompt Engineering eXpert) addresses a key limitation in current prompt optimization methods: inefficient use of evaluation budgets due to static dataset treatment. The framework introduces dynamic data stratification, categorizing examples into Easy, Hard, and Mixed performance tiers based on optimization history. By focusing on the Mixed tier—where language models demonstrate inconsistent performance—APEX identifies two high-leverage subsets: the addressable frontier for generating informative prompt mutations and the rank-sensitive frontier for distinguishing between candidate prompts. The approach was evaluated on three benchmarks: IFBench, SimpleQA Verified, and FACTS Grounding. Results demonstrate that under a constrained budget of 5,000 evaluation calls, APEX substantially outperforms baseline prompts, with improvements of 11.2% on Gemini 2.5 Flash and 6.8% on Gemma 3 27B, indicating that data-centric optimization strategies may be more effective than existing evolutionary algorithm paradigms.
What's missing
The paper does not discuss computational costs of the dynamic stratification process itself, potential limitations when applied to other model architectures or domains beyond the three tested benchmarks, or how performance scales with different budget constraints. Additionally, the study does not compare against other data-selection strategies or provide analysis of failure cases.
What different sources said
- arXiv cs.AICenter
APEX: Automated Prompt Engineering eXpert with Dynamic Data Selection
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.