Please Let Me Read – The Web Was Once Good:( | Hacker News Reader

Please Let Me Read – The Web Was Once Good:(

github.com

18 pointssubbu_devhub25 comments

I just can't enjoy reading something so obviously AI generated. Write it yourself.

@PosedProto I did use AI to generate the README it serves the purpose, do you feel writing it myself would have made a difference, I am not good with words so I Took help where I could

I will try to write it myself and update it, thanks for feedback!

I feel the author could have applied the declutter tool to their readme.

Or perhaps the LLM should have known to do that.

what about it would you change, specific feedback is welcome * remove emoji ? * layout much better

I am dyslexic so have a hard time arranging things, so I gave the basic idea and the AI generated the README

I've been using a webapp I build to do the same thing, and I've really enjoyed it.

https://yazzy.carter.works/

Example: https://yazzy.carter.works/https://paulgraham.com/submarine....

https://github.com/carterworks/yazzy

It uses Steph Ango's/kepano's defuddle under the hood.

The live version is hosted on a single free-tier fly.io node, but it is easily self-hosted.

this looks great @open-paren will check it out, for some reason the webapp take a while to load I can take it for a spin locally though

So far I've found Firefox reader mode does what this aims to sufficiently well if I plan to read the page online, and saving a minimal version to Joplin works if I want it offline.

@Zak_2 I tried the reader mode in several browsers but it just did not work! there were several issues and sometimes the majority of the content was removed, it was such a hit or miss

My idea is admittedly half-baked, but I've been thinking of the concept of an alternative web. I don't know how it would work. Ignoring implementation for a second, what if you could stay within a network of non-corporate content, search engines, etc.? I feel like everyday I find neat websites on Hacker News built by real people for fun and community, but then my daily web browsing activity doesn't feel like this. If I do one Google or DuckDuckGo search, I'm inevitably going to land on some site that is at least cluttered with SEO garbage, even if it's an individual person's blog (cooking blogs are a great example). I guess my criteria for what sites would get included in this are not well defined. Maybe I just want the old web back.

So like: https://geminiprotocol.net/

They claim the protocol is resilient to enshittification.

Does anything really need to be built for that? From the first moment, you were only interested in a fraction of the Web. There has been a constant arms race between the search engines and the optimizers. We have always relied on suggestions from friends to sort out the "good" content (for whatever definition of "good").

There's much more total stuff now, and probably the fraction of "good" stuff is much smaller. But the total amount of it as at least as high.

I agree with the premise and don't want to knock someone's project.

It does seem like a lot of computational effort to achieve what F9 / Reader View does in FF.

You could be right. But plenty of websites with useful information intentionally obfuscate their pages to trick the browser into disabling reader mode.

@focusedone I tried reader mode in several browsers it was such a hit or miss, it just did not work for me, and honestly I wanted to convert to markdown not just plain text

I tried several reader modes, there were several issues including * several potions of the main content was missing * the navigation bits get caught when in reader mode * the comments and other un-related sections come in play

I really tried these before invesitng time in this

Oh, cool! How does this do with the intentionally obfuscated sites?

@focusedone if you see the code link here https://github.com/subranag/declutter/blob/main/src/page.ts there are some specific techniques recommended to simulate normal browsing behaviour, but it does not work 100% percent of the times, but works on most of the sites

for example * simulate scrolling after page loads * simulate plugins * simulate location etc

once all of this is done hopefully the HTML content becomes readable

README oozes ai generated copy.

Does this do anything that Firefox Reader Mode doesn't? Just seems like a non-deterministic, slower, non-free version.

Also, I understand writing READMEs is boring, but please at least edit what the LLM produces. You do not need this many content-free emoji bullet point sections.

Edit: Looking at the prompt made me realize that the output of this would obviously be completely untrustworthy: https://github.com/subranag/declutter/blob/main/src/llm.ts#L...

@burkaman why would you think this would completely untrustworthy, I Can include some sort of content matching heuristic to make sure that there is some level of match.

why is everyone against emojis :) I personally feel they are fun!

I Am dyslexic so I have a bunch of ideas and I have a hard time laying that out in coherent text, so I gave the basic idea to the LLM and it wrote the README in the structure I want.

but feedback taken I will try to write it myself, thanks for taking time out to provide feedback!

Nah. Hard pass on even more AI inbetween me and the few bits of original content that are still out there. Reader mode and we're good.

Or use Safari’s reader mode, it works well!

For the love of God, people need to get back to writing their own Readmes and not just taking LLM output unedited. I do want to read, but I don't want to read ChatGPTese.

Sorry, let me ask ChatGPT to put it in terms people seem to prefer now (I don't think this stuff is actually quite right but who cares anymore):

    ## 1. They Optimize for Politeness, Not Usefulness
    
    ChatGPT READMEs tend to:
    - Over-explain obvious things
    - Avoid strong claims
    - Hedge unnecessarily  
    The result is text that feels safe but not informative. A good README should reduce uncertainty quickly, not pad it with disclaimers and filler.
    ## 2. They Follow Templates Instead of Intent
    Most generated READMEs look structurally correct but contextually shallow:
    - Generic section headings (“Installation”, “Usage”, “Contributing”) regardless of relevance
    - Boilerplate language that could apply to almost any project
    - No clear prioritization of what actually matters
    This signals that the README was assembled, not written with purpose.
    
    ## Summary
    ChatGPT READMEs are usually:
    - Correct but unhelpful
    - Polished but shallow
    - Complete but low-signal

@roywiggins I have a hard time writing a wall of text, I am dyslexic so I have a bunch of ideas and I give the rough outline and the LLM generates the doc that looks fairly polished so I use it where ever I can

I will take your feedback and see if I can restructure it better! thanks for taking time out to structure your comment properly, gives me a good insight on how to write good READMEs

I think you would be better served by a shorter, human-driven README hitting the important points and then a separate AI summary doc that's marked as such. Your rough outline is what I'd probably prefer to read.

  The modern web is broken. Before you can read anything, you hit a wall: popups,
  ads, paywalls, tracking scripts, and navigation clutter designed to keep you clicking
  instead of reading.

  Declutter strips all of it away using AI. But here's the thing, you don't need
  frontier models. Even small, cheap models like Gemini Flash or Claude Haiku do an
  excellent job extracting pure content. That means you can archive whatever you want
  to read for pennies per month.

  Just point it at a URL, and out comes beautifully formatted Markdown, HTML, or PDF.
  Offline. Local. Clean.

  I built this because I was tired of the fight. The web doesn't have to be this way.
  I just want to read. Please let me read without distractions.

Reply on news.ycombinator.com