My Process

Some people have asked me how I produce the content for this Substack.

Here’s my process.

1. I use the search filters on the Science Open website to search for research papers in the following disciplines that were published during the previous month.

2. Altmetric is a service that ranks research papers by the number of mentions across social media, news publications, and research journals. I sort the results by the Altmetric score to surface (what I hope) are the most relevant, influential, interesting papers.

3. I then select between 8 and 15 of the top papers, take the often very technical abstract, and then prompt Claude AI to translate it into language that my and my peers can better understand.
   
   Prompt engineering is very much a process of trial and error. However, after a lot of experimentation (approx. 35 different prompt approaches), I was able to come up with a prompt that nicely translated most paper abstracts into language that makes sense to me.

   
   Note: Claude AI is a large language model similar to Chat GPT but with more recent context and a larger context window.


4. When Claude AI has translated the paper abstract into a more simple language, I read through it to see if I can understand the purpose of the research, the testing methods used, the outcomes and the implications and ramifications of the research paper.
   
   Note: If I don’t understand the simplified abstract, I prompt Claude AI repeatedly until it makes sense to me and, therefore, my peers.

5. That done, I collect all the translated summaries together and post them into a new Substack post for the current month.

6. My father is an AI developer, and without his guidance and technical support, I could never have made my idea come to fruition.

   
   P.S. I use Midjourney AI to create the images.