Hacker News Reader

Comment by hansword | Hacker News Reader

The way this is described it reminds me of the <details> and <summary> tags in HTML, most often used for spoilers.

But I am not sure if I understood it correctly. This might be more for different level of comprehension (eli5?).

I could be wrong, but I'm interpreting it as a sort of analogous system to Word's autosummary feature. That is, You put in a complete text, and then have periodically briefer summaries of it (perhaps in sections, such as paragraphs?). You start with the summarised version, but then as the reader stretches out the text, it dynamically provides more detail in the areas that are stretched out.

e: For those like me who just went looking, apparently autosummary was removed in Word 2010, which given I had 2007 til about 2 years ago and thus didn't realise it was missing :)

Here's a description of it: https://www.officetooltips.com/word_2007/tips/getting_to_the...

I'm as baffled as you. I did find this that may or may not help. It only very barely helped me. https://ia904500.us.archive.org/32/items/STRETCHTEXT/STRETCH...

It sounds like you have to write multiple versions of the text, and then if the user grabs one of the markers and "stretches" it, the next-most detailed version of the text will get loaded.

But I could be very wrong. It is extraordinarily difficult to find something concrete as a demo. I gave up on link after link of the same description with no actual demo anywhere.

> It is extraordinarily difficult to find something concrete as a demo.

I think that's because it would be extraordinarily difficult to write something good using StretchText. It would amazing if it could be done, but the amount of effort needed to do so is probably the main reason why no-one does.

With the advances in machine learning, I wonder if you couldn't automate it. Write the most detailed level and have some ML model generate the ever more high-level summaries.

Yes, exactly. This is also applicable to building good prompts for language models from larger document sets.

Here is an example of using Github page summaries to seed knowledge in discussing various topics with GPT-3: https://github.com/daveshap/LongtermChatExternalSources

My brother-in-law built a very small demo to advertise an implementation of StretchText he made years ago:

https://natematias.com/stretchtext/

I don't recall whether he ever built anything larger with that.

Reply on news.ycombinator.com