English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Generalist foundation models such as GPT-4 have displayed surprisingcapabilities in a wide variety of domains and tasks. Yet, there is a prevalentassumption that they cannot match specialist capabilities of fine-tuned models.For example, most explorations to date on medical competency benchmarks haveleveraged domain-specific training, as exemplified by efforts on BioGPT andMed-PaLM. We build on a prior study of GPT-4's capabilities on medicalchallenge benchmarks in the absence of special training. Rather than usingsimple prompting to highlight the model's out-of-the-box capabilities, weperform a systematic exploration of prompt engineering. We find that promptinginnovation can unlock deeper specialist capabilities and show that GPT-4 easilytops prior leading results for medical benchmarks. The prompting methods weexplore are general purpose, and make no specific use of domain expertise,removing the need for expert-curated content. Our experimental design carefullycontrols for overfitting during the prompt engineering process. We introduceMedprompt, based on a composition of several prompting strategies. WithMedprompt, GPT-4 achieves state-of-the-art results on all nine of the benchmarkdatasets in the MultiMedQA suite. The method outperforms leading specialistmodels such as Med-PaLM 2 by a significant margin with an order of magnitudefewer calls to the model. Steering GPT-4 with Medprompt achieves a 27%reduction in error rate on the MedQA dataset over the best methods to dateachieved with specialist models and surpasses a score of 90% for the firsttime. Beyond medical problems, we show the power of Medprompt to generalize toother domains and provide evidence for the broad applicability of the approachvia studies of the strategy on exams in electrical engineering, machinelearning, philosophy, accounting, law, nursing, and clinical psychology.

Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine