All right, if you were looking for a page containing the serious stuff, you have found it.

What is NeurExplain?

NeurExplain is an online initiative to use generative AI (for starters, GPT-4) to enhance scientific abstracts (currently retreived from PubMed). In the long term future the goal is to move towards AI-driven graphic medicine. All artwork on NeurExplain has actually been created by DALL-E3.

PubMed search

For the abstracts on this website, we developed an algorithm to query PubMed from a predefined set of MeSH terms, picking one at a time. This step does not involve generative AI. The sort order is “most recent”. We understand that some abstracts do not seem related to the category under which they are classified, but this is what PubMed returns. For example, check the results on cardiology.

Does GPT-4 provide reliable information?

GPT-4 scored an overall accuracy score of 85% on USMLE medical reasoning and 90% on the USMLE soft skills, significantly outperforming the freely available ChatGPT (based on GPT-3.5) which “only” scores 53% on reasoning and 62.5% on soft skills. It is definitely not perfect, but what human doctor is? Decide for yourself whether you’re willing to bet on a 85%-90% USMLE score!

Disclaimer

As you may have guessed, despite optimal training of the bots we cannot guarantee that all interpretations of the abstracts are correct. So before applying, check for yourself (e.g. using the DOI or PMID with the article).

Privacy stuff

To the best of our knowledge we are not collecting your personal data here. The site is built on WordPress.org and uses these plugins: Akismet, AIOSEO, Limit Login Attempts, Mailchimp for WordPress, Sassy Social Share, and WPForms. The server is hosted with Cloud86 and located in Europe. The mailing list is based on MailChimp, and our Python scripts run on the EU servers of PythonAnywhere.

That’s it. Definitely reach out if you think there are privacy issues, because then we’d like to know!