Me discussing the article with my family:
Going to bed now but I will add it’s rather long and the good bits are in the footnotes
It’s less of a must-read than a research project for follow up — very good for ‘situation awareness’
So I went through https://ai-2027.com and extracted and annotated the links I wanted (what we called in the 20th century ‘making an annotated bibliography’). I did this in the form of a markdown table.
I then took the markdown table, pasted it in https://dillinger.io and exported a pdf. However, I wanted to know how many rows were in the table and it was 9 pp. long (the answer is 80). Also, I wanted it in RDF format so I could import it into a knowledge graph and query it with the SPARQL language. I asked Claude 3.7 (‘sonnet’) to do this for me:

Labour involved:
- several hours of ‘human intelligence tasks’ (HIT, like Mechanical Turk) of clicking on about 180+ footnotes, capturing any links, and pasting those links into a table in a markdown editor)
- a few minutes to generate the PDF at a website that does that.
- a few minutes monitoring Claude doing the conversion (took a couple tries), a.k.a. vibe coding
- taking screen shots and ‘live blogging’ the experience.
Leave a Reply