Thursday, June 1, 2023

Error converting content: marked is not a function

location:: Home - Avalon, Boston 
victory-hour::
pomodoro-blocks::
wake-up:: 
day::
icon:: 📅🗓⚡️🌟🌟🌟🌟
  - **21:48 PM, day. Morning Pages**
- Late here but here. Hello second brain.
- BASB called me back to Circle App. Found Book Summary article there. #justintime
  collapsed:: true
- @Tiago Forte on his attempt to summarize book via ChatGPT
  - Issue with just asking chatgpt for a summary
  - **Overuses clichés:** It uses a number of clichés that don’t really communicate anything, such as “thought-provoking” and “balanced approach”
			  id:: 64794ac4-029e-43bc-8ce2-230ff317b1ae
  - **Boring, conventional language:** I remember that much of the value of the book was in restating commonly known truths in new and surprising ways, yet ChatGPT used conventional, rather boring language throughout (such as “relentless pursuit of efficiency,” “improved decision-making,” and others)
  - **Lack of specifics:** There’s also an unfortunate lack of detail, with ChatGPT’s summary hinting at “several case studies and examples” without mentioning any
  - then trying elaborate prompt
  - **Pandering, patronizing language:** Often found in parenting books, this kind of language tries to make parenting sound “cool” but comes across as annoying (“symphony of wailing children”)
  - **More clichés:** It replaced the clichés of business writing with the clichés of self-help, which are even worse (“emerge as a creative powerhouse, brimming with inspiration and purpose”)
  - **Generic examples:** The examples it cites are incredibly generic and unhelpful (“the power of ‘serendipitous collisions’”)
  - - "And of course, by this point I’ve already done most of the work of summarization that I was trying to avoid in the first place!" #lol
  - I’ve arrived at a decent summary that is perhaps 50% as good as one I could have created myself, while saving perhaps only 20-30% of the total effort needed to get there.
  - Conclusion
  - First, I concluded that supplying ChatGPT with the best excerpts from one’s reading greatly improves the summary it can provide, especially when those excerpts are organized into an outline that provides a particular point of view.
  - The main challenge is distilling the excerpts enough to fit within the context window (though upcoming improvements to ChatGPT include a vastly larger window that will fit more than 15 times as much text).
  - Second, you can’t instruct ChatGPT to come up with “surprising” or “counterintuitive” ways of explaining things, because those qualities depend on sensitivity to a cultural zeitgeist made up of millions of people’s joint expectations, plus access to a wealth of details to put a fine point on its assertions. Instead of *being* surprising, it will merely use words like “surprising” and “counterintuitive” in its otherwise boring description.
  - Third, ChatGPT is always tending toward cliché, which by definition is the “aggregate average” of the greatest number of writers on which it was trained. You have to constantly come up with inventive ways of getting it off that well-worn path.
  - Fourth, I conclude that reading books is still worthwhile, even in an age of Artificial Intelligence. Besides the personal enrichment from the experience of reading, there are key details, subtle distinctions, and gestalts of meaning that come from reading and notetaking that can’t (yet) be reproduced by even the latest language model
  - -