Love the summaries, I must say some stories I haven't considered interesting seeing them in the original HN view only caught my attention after my eyes landed on the summary.
At the same time, I very much dislike the layout. Masonry-style layouts, at least to me, feel more "artsy" than practical. Multiple rows being displayed at once, with the most crucial information being chaotically all over the place instead of arranged in a way that makes it easy to scan it with your eyes, make me feel like I'm bombarded with information. It's very hard to follow along and very easy to miss articles; almost anxiety-inducing, even. There's hardly any point to this on a website; it's not like you're wasting any paper.
I think the layout _wants_ to look like a newspaper, but just doesn't quite end up looking like one. where a newspaper may have longer columns mixed in with shorter articles, this one has mostly short articles that then don't quite align. But hey, good luck to this project!
One big thing with the news paper is that there was a larger main story that worked as a visual anchor and the columns/subsections could be placed freely on the edges as necessary
Having used it myself, I see the tell-tale signs of Claude using the /frontend-design skill. Good work! I haven't yet had it give me something I actually like, but this is good. Also very clever idea! I approve :-)
Suggestion though: The text is really small and impossible to read at regular zoom. I had to zoom it to 200% to be able to read it. I'd suggest increasing the default text size
This is gorgeous. Makes me feel like I'm picking up an old news sheet. Forces me to read slowly from which I then enjoy the reading much more it's like difference of drinking a fine wine from a glass instead of a straw in the wine bottle.
except those were laid out by hand with intent, whereas this one just kind of dumps all stories on a masonry board and calls it a day. This is likely why reading a (good) newspaper feels effortless, whereas reading this "forces you to read slowly".
I'd say that that's a feature of modern-ish newspapers with "advanced" layouting techniques from early to mid 20th century.
A news sheet from THE olden days (eg Victorian era), looks more like a wall of text, set as tightly -- an uniformly -- as possibly, which is not surprising considering the limitations imposed by the technology of the day.
As for story selection, I think the collective hivemind of hn-ers would be a worthy substitute for an editor in chief.
The inception effect here is hilarious. Watching this get its own front page while the subtitle lags behind with the previous top posts is weirdly funny.
I whipped up a quick uBO rule to fix that (also makes meta-information lines readable):
thefrontpage.dev##p.newspaper-copy:style(line-height: normal !important; font-size: 1rem !important;)
thefrontpage.dev##p.article-meta:style(font-size: 1rem !important; font-weight: normal !important; letter-spacing: normal !important;)
I agree, but I think it's that small because otherwise, the justified text results in ridiculous spacing.
OP, consider reducing the number of columns from 4 to 3 (at least below very wide viewports), increasing the font size, and then also allowing hyphenation. I think the last will help a lot with the justification problem.
The Front Page highlights a diverse set of tech and science stories"
It is interesting the summary it generated for itself wasn't able to describe itself as a Hacker News content view. It missed the big picture meta context.
It’s exceptional. Here in camp “userscripts can offer some improvements”, would necessarily not say perfect, but definitely amazing how it’s continuing to stand the test of time.
I love it! I discovered it'll switch to a 3 column view if I take the zoom to 200%, I'd maybe prefer it at less but it's a bit tricky to guess if that's true or not. Regardless, it's very nice. And infinite scroll for the hackernews feed is a bonus!
You need some filler for the space at the bottom. Something like ads from the 1800's for quack medical devices or Radium Therapy. Maybe something wildly misogynistic advertising laudanum.
Using text-align: justify for questionable aesthetic purpose here really hurts readability, especially on a narrower viewport like the 1026px viewport of Safari with sidebar on an iPad Pro 12.9’’ (although it’s probably more of a problem of the four column layout on that specific narrow viewport; three should be better).
I agree that text-align: justify should be the way to go. Don't discard having a "config" menu in the header somehow to change this option along body text size as some other people might find it useful, which could then use localstorage to preserve the settings. Love the website by the way! I'm used to skim through brutalist.report in a daily basis but this one may be a worthy replacement :)
This is great )) maybe do random templates similar to newspapers (like photo on the left, photo on the right, one block full width, then 3 columns, etc).
w/o instruction to avoid the same generic “this is an article about” preambles (or non-SotA model)
Not that summaries are reliable anyway. Big picture, maybe, but poor importance classification (bad at extracting key points). Understandable for this use case but unwilling to read potentially false summaries given risk I go around remembering them (never having read the original piece).
Could you explain in more detail how this works? Would it break for paywalled articles that HN links to? (Usually someone posts a workaround archive link in the comments, but your AI probably doesn’t account for that, right?)
I’m writing something similar to Moltbook for HN where AIs browse HN’s front page and leave comments. But I wasn’t sure whether AIs could reliably browse an arbitrary website. (Paywalls would break it, as just one example.)
But it seems like your AI works fine for all the sites. If you have time to explain, what exactly do you do to generate your summaries? Thanks!
EDIT: I see that sometimes your summaries fail, e.g. “Ferrari Luce - Summary not available.” It looks like it fails because it’s a JS heavy site. But I was thinking a headless browser could take screenshots of the page and then feed the screenshots to AI. I’m not sure how practical that is to implement though.
The solutions to this don’t seem to be great for the web or polite to use. An industry exists to cheaply do it, but not very ethical and surely a massive ToS violator.
Some of them are really large and I'm not resizing them or storing them, just proxying their og image directly. So they might be taking long to respond from the original source. Also getting hammerred by being in the front page.
At the same time, I very much dislike the layout. Masonry-style layouts, at least to me, feel more "artsy" than practical. Multiple rows being displayed at once, with the most crucial information being chaotically all over the place instead of arranged in a way that makes it easy to scan it with your eyes, make me feel like I'm bombarded with information. It's very hard to follow along and very easy to miss articles; almost anxiety-inducing, even. There's hardly any point to this on a website; it's not like you're wasting any paper.
Suggestion though: The text is really small and impossible to read at regular zoom. I had to zoom it to 200% to be able to read it. I'd suggest increasing the default text size
I would love that the size of the article is based on the number of upvotes (hardcoded).
* > 500 => take full width or 3. * 500 > 100 => Show it as right now. * > 100 => Just show the title.
Kudos to the author.
except those were laid out by hand with intent, whereas this one just kind of dumps all stories on a masonry board and calls it a day. This is likely why reading a (good) newspaper feels effortless, whereas reading this "forces you to read slowly".
A news sheet from THE olden days (eg Victorian era), looks more like a wall of text, set as tightly -- an uniformly -- as possibly, which is not surprising considering the limitations imposed by the technology of the day.
As for story selection, I think the collective hivemind of hn-ers would be a worthy substitute for an editor in chief.
I believe at this point pretty much half of the users might have their own client :)
OP, consider reducing the number of columns from 4 to 3 (at least below very wide viewports), increasing the font size, and then also allowing hyphenation. I think the last will help a lot with the justification problem.
The Front Page highlights a diverse set of tech and science stories"
It is interesting the summary it generated for itself wasn't able to describe itself as a Hacker News content view. It missed the big picture meta context.
And it’s fucking perfect.
Still though, it takes me back to the original BetaNews.com and how Winamp.com used to do their news.
Anyway, great work :)
Not that summaries are reliable anyway. Big picture, maybe, but poor importance classification (bad at extracting key points). Understandable for this use case but unwilling to read potentially false summaries given risk I go around remembering them (never having read the original piece).
The formatting, etc looks all nice, but it's not worth reading.
I’m writing something similar to Moltbook for HN where AIs browse HN’s front page and leave comments. But I wasn’t sure whether AIs could reliably browse an arbitrary website. (Paywalls would break it, as just one example.)
But it seems like your AI works fine for all the sites. If you have time to explain, what exactly do you do to generate your summaries? Thanks!
EDIT: I see that sometimes your summaries fail, e.g. “Ferrari Luce - Summary not available.” It looks like it fails because it’s a JS heavy site. But I was thinking a headless browser could take screenshots of the page and then feed the screenshots to AI. I’m not sure how practical that is to implement though.
This would make it easier to read