Personalized digest#27

Rather than me going and reading through articles suggested by Scour, I would be curious what an LLM-generated digest would look like.

Here are my thoughts:

• Instead of a boring system prompt like “summarize these articles”, let’s make it more personal, for example: “You are @justmoon’s work colleague and friend and you are obsessed with following the latest tech news. You know that @justmoon cares about these interests: [interests]. Read the following posts carefully and think about which ones @justmoon may find interesting and why.”

• The focus should be on surfacing (scouring for) hidden nuggets that might otherwise get lost in the noise, not just summarizing the “most important” news.

• The digest would not be intended to replace reading the articles but merely providing a more streamlined discovery experience.

• Therefore, the digest should focus on the “why it’s interesting” and not the “what it contains”

• Ideally, the assistant would know roughly what I know and don’t know. For example, it would know my proficiency level with different technologies.

• You can think of this as a second processing layer on top of the basic relevance layer. I.e. the current Scour is there to narrow down the millions of articles written every day to a few hundred and then this step narrows it down to the top ten that I’m actually going to read.

That’s a great suggestion! I’ll play around with something like that and see what it looks like