{"slug": "can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of", "title": "Can Large Language Models Optimize Pretrained Neural Network Weights? A Multi-Phase Empirical Evaluation of Gradient-Free Editing in YOLO26 Object Detection", "summary": "A multi-phase arXiv study (2611.17789v1) testing whether large language models can edit pretrained YOLO26 object-detection weights without gradients found no statistically established LLM advantage over random-policy guidance, with p-values of 0.612, 0.369 and 0.694 at budgets of 8, 25 and 50 evaluations. Raising the CMA-ES budget from 8 to 250 evaluations lifted development gains from +0.11 to +0.28 percentage points on 1,000 COCO Val2017 images, but selection changes stayed negative (-0.03 to -0.12 points), and a five-seed YOLO26m replication produced a COCO API AP difference of just +0.004 points (95% CI: -0.026 to +0.034). Of 225 archived LLM proposals, 53 met the final strict schema, and the authors conclude genuine LLM-mediated weight editing is feasible but the tested approaches showed no consistent transfer beyond the development subset.", "body_md": "arXiv:2611.17789v1 Announce Type: new \nAbstract: Direct weight editing offers a gradient-free alternative to fine-tuning pretrained detectors, but its generalization and the value of large language model (LLM) guidance remain uncertain. We evaluate five experimental phases using YOLO26 models and MS COCO, comparing deterministic scaling, random perturbations, differential evolution, CMA-ES, structured interventions, real local-LLM proposals, failure-aware guidance, and hybrid LLM-CMA-ES against a matched random-policy control. Optimization uses native Ultralytics mAP50-95 on 1,000 COCO Val2017 images; selected candidates are assessed on a disjoint, previously examined 1,500-image subset, with COCO API AP50-95 reported separately where available. Increasing the CMA-ES budget from 8 to 250 evaluations raised development gains from +0.11 to +0.28 percentage points, while selection changes remained negative (-0.03 to -0.12 points); the 250-evaluation result uses one seed. The hybrid LLM approach showed no statistically established advantage over random-policy guidance (p = 0.612, 0.369, and 0.694 at budgets 8, 25, and 50). Five-seed YOLO26m replication yielded a COCO API AP difference of +0.004 points (95% CI: -0.026 to +0.034). Of 225 archived LLM proposals, 53 satisfy the final strict schema; early tensor-level execution cannot be fully verified. Genuine LLM-mediated weight editing is feasible, but the tested approaches did not demonstrate consistent transfer beyond the development subset or a measurable LLM advantage.", "url": "https://wpnews.pro/news/can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of", "canonical_source": "https://arxiv.org/abs/2611.17789", "published_at": "2026-11-16 05:00:00+00:00", "updated_at": "2026-11-16 05:21:35.876184+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "computer-vision", "ai-research"], "entities": ["YOLO26", "MS COCO", "Ultralytics", "CMA-ES", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of", "markdown": "https://wpnews.pro/news/can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of.md", "text": "https://wpnews.pro/news/can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of.txt", "jsonld": "https://wpnews.pro/news/can-large-language-models-optimize-pretrained-neural-network-weights-a-multi-of.jsonld"}}