Skip to main content
Search

# Prompt Engineering

10 related articles

Subscribe to updates
  1. aclanthology.org

    SELFHEAL cleans poisoned fine-tuning data before training: up to 85% lower attack success on Llama-3, with natural accuracy held

    SELFHEAL moves the defense ahead of fine-tuning: instead of discarding samples, it uses only the base model's own language judgment to tag instructions, repair grammar and coherence, and rewrite suspicious responses. On Llama-3 it cuts attack success by up to 85% relative to post-hoc defenses while keeping natural accuracy comparable to a model trained on clean data; on Mistral the effect weakens markedly, showing the cleansing depends on the base model's own text understanding.

  2. Fireworks AI

    Evals stop at failure reports: GEPA writes them into the prompt, lifting Qwen3-32b Text2SQL test accuracy by 10%

    Eval Protocol plugs GEPA prompt optimization into an existing evaluation interface, so the same datasets, metrics and task constraints both measure an agent and automatically rewrite its system prompt. In a Text2SQL case, Qwen3-32b gained 10% test-set accuracy and more than 5% validation-set accuracy without touching model weights, with about 45 minutes of prompt search; the same eval then served as the objective for RFT.

  3. More detailed prompts did not yield better helmet classifiers: the most prescriptive tier dropped two LLMs to about 0.58 accuracy

    ChatGPT, Claude, Gemini and Grok each generated PyTorch helmet-compliance training scripts from three prompt tiers of increasing specificity, run 20 times on 972 balanced images: ten of twelve configurations converged to a 0.957–0.967 accuracy plateau, but the most detailed tier left Gemini and Grok near 0.58, with scripts that ran yet failed to converge. A backbone-by-scheduler ablation traced the failure to the prescribed from-scratch multi-scale network, which collapsed under both schedulers while a pretrained ResNet18 converged under either.

  4. Rules made six frontier LLMs ignore the scoreboard; two prompt sentences brought them back

    Across 2,160 sessions of a modified rock-paper-scissors game, six frontier LLMs became insensitive to changing outcomes once a rule was supplied, and two sentences of direct tracking induction in the system prompt significantly reduced that insensitivity in all six.

  5. medRxiv

    Uncoded Clinical Features from Multilingual Electronic Health Records in Catalonia: Development and Validation Study

    This study deployed the 3.8B-parameter open-weight small language model Phi4-mini (via Ollama) within an institutional firewall, combined with deterministic regular expression post-processing, to extract six uncoded urinary tract infection clinical features (fever, nitrites, leukocytes, lumbar pain, abdominal pain, and haematuria) from 15,498 Catalan/Spanish bilingual MEAP primary care narratives in the SIDIAP database in Catalonia, achieving 93.8% accuracy, 96.6% specificity, and 83.6% sensitivity in a double-blind clinician gold standard validation of 60 real patient records, and 87.2% accuracy, 99.5% specificity, and 99.2% positive predictive value in adversarial synthetic stress-testing of 720 notes, with no patient data leaving institutional servers.

  6. Harnessing Generative Artificial Intelligence and a Persona Prompt to Develop Competence in Addressing Social Determinants of Health

    A family nurse practitioner (FNP) program had 98 students use a generative AI persona prompt in Claude Sonnet to create a virtual patient on a video-conferencing platform, and after a 30-minute prebriefing, about 75 minutes of scenario work, and a 45-minute PEARLS-based debriefing, the most common one-word reaction was “helpful” (n = 83), 95% said the activity facilitated their learning about SDOH (n = 82), 78% chose the two highest comfort levels for screening for and addressing SDOH (average 4.1/5), the SDOH quiz average was 94% (n = 98), and 54% answered the question on specific strategies to address SDOH correctly.

  7. Multimodal Large Language Models for Bladder Tumor Detection in Cystoscopy: A Retrospective Benchmarking Study

    This retrospective study analyzed 1,754 labeled public cystoscopy images to test Direct, Book-based, and Optimized prompts across GPT-5.2, GPT-5, GPT-5-Mini, and GPT-5-Nano for benign-versus-malignant classification, finding that GPT-5 and GPT-5-Mini with the optimized prompt reached accuracies of 86.7% and 89.2%, and that GPT-5 with the optimized prompt achieved 98.1% accuracy, 94.6% specificity, and 99.1% sensitivity in high-confidence triage at 62.0% image coverage, while prompt engineering improved calibration and triage without statistically significant performance gains.

  8. Optimizing Large Language Models for Hospital Discharge Prediction: A Retrospective Cohort Study with Inference-Time Optimization Strategies

    In this retrospective cohort study at a single tertiary academic medical center, large language models predicted same-day discharge from clinical documentation in the 30 hours preceding a 06:00 index time for adult inpatients admitted in 2024 with a length of stay between 2 and 14 days, using a randomly generated validation set (n = 860) and test set (n = 886); the baseline GPT-5 prompt achieved an F1 score of 0.48 and sensitivity of 0.

  9. medRxiv

    An LLM-Enabled Pipeline for Natural History Study Information Extraction in Rare Disease Research

    This study built a proof-of-concept information extraction pipeline using three open-source LLMs (Athena-v3-AWQ, Gemma3-27B, and Llama-3.1-70B-Instruct) to extract 11 natural history study characteristics from PubMed abstracts labeled "2" in the CZI DRSM corpus (148 gold-standard and 3,547 full-corpus abstracts), finding that all three models exceeded 99% processing success, that Gemma was best overall on expert rating (68.0% of outputs rated "good") and full-corpus runtime (~16 minutes), that Llama scored higher on automated Token F1 (0.874 vs. 0.723), and that Athena performed worst largely because it copied source text verbatim rather than synthesizing it.

  10. Prompt Engineering in the Segment Anything Model: Methodologies, Applications, and Emerging Challenges

    This survey systematically reviews prompt engineering research for SAM and its growing ecosystem, proposing a hierarchical taxonomy that organizes methods into geometric prompts, textual semantic prompts, and multimodal fusion prompts, further tracing the transition from manually crafted prompts to automated generation based on detector outputs, prototype learning, reinforcement learning, and vision-language models, while tracing how prompt engineering enables cross-domain generalization in medical imaging, remote sensing, industrial inspection, and anomaly detection, and identifying key challenges such as prompt sensitivity, cross-modal misalignment, and computational inefficiency alongside future directions including causal prompt reasoning, collaborative multi-agent prompting, and diffu

You’re all caught up