While others have been working diligently to destroy reading experiences using pretext, I’ve been thinking a bit about how to improve them using llms. Here’s a tldr concept where you can seamlessly swipe between synopsis and original article while maintaining your reading position at all times.

ok this is actually cool man, allow me to steal the idea for future projects

Ya man, keep it going👌

Yo! I had a similar idea for this, except for whole books, where you generate 3-4 levels of granularity of summarities that you can navigate between Question is tho how do you map between point in (condensed) summary to point in actual text?

Yess, could work for whole books too. Yep, that indeed is the tricky part. I segmented into chunks around 2k chars each then did 2 summaries for each segment, one high compression, one low compression. Then while reading, basically track the progress through the current segment and do linear interpolation to find equivalent char offset in other variants of that seg. It’s not 100% accurate but if you keep the segment size small enough it’s accurate enough.

Ahh very cool. Might use this trick for my personal blog. Cool work!

Thanks, you should!

Always with the best demos! 💯

Nice!! Is this using Pretext or no?

nah, just markdown renderer & framer motion

Awesome! built something similar a bit ago Though i've realized that traditional reading, listening, or pasting-in-gemini-and-asking-questions are the best ways for me to consume essays

Nice, cool! I like the side-by-side too. Hoping I can hook it up to our tts and then just listen at 1.5x speed for max laziness haha.

yes, tts is awesome! recently made a simple swift app to narrate arbitrary content using Google's free tts models though I'm sure ElevenLabs would do even better (and could narrate in the author's voice)

Perfect for specs (e.g. plan.md), the switch exactly makes specs much more readable. Most specs are way too long, and almost nobody reads them fully, same story happens in traditional tech design docs review meeting, nobody really reads full text.

True! Good use case

lol, little jab to all the pretext posts but valid though. I appreciate good UX.

Lol, just a good opportunity for a little cheeky sarcasm. Obvs i appreciate all that work and that we're bringing some fun elements to the web again✨

Anything to nudge the algorithm;)

very cool dude

cleaner version of bionic reading

Neat.

wow this is very cool!

@mannupaaji Great work

is this OS?

It’s nott

this is great! wondering a simple progressive blur instead of the linear fades at the top and bottom of the page might be a nice subtle touch to add ✨

wow this looks promising

Genius tbh - swiping through the modes on mobile would be sooo nice for so many articles I read

See also @prathyvsh's for a great collection

This is a smart concept

@mannupaaji This should be useful for epub.js

this is way more intresting than another chatbot wrapper improving how people read beats building another tool nobody asked for

if there's a 🔗 i will share it

Nothing public atm

This is actually really crisp, I would use this.

Thank you for this care❤️

Seems a little odd that the single page displays x out x pages. Are those chapters? The paragraph reveal appears to be inverted - ideally the top part softens out while the bottom stays at full contrast for a continuous read rather than actively revealing

Since people are accustomed to reading books, it represents roughly the size of a page in a book. The ux is designed to force the user to read on the center line because that’s the only position where alignment is guaranteed when transitioning between variants. Different variants have different information densities, so as your focus shifts from the center, alignment will slowly be lost and you’ll lose your reading position when switching variants.

I appreciate this and I can see where you're coming from. I read a lot in all forms. So I'm pretty surprised to learn there are people who might want to read this way. And I don't doubt you'll find some good use for it