Kagi News • Hacker News Board

Re: Kagi News

#211

Post by Harmon758 » Tue, Sep 30, 2025, 6:23 PM UTC

__jonas wrote: post #23

Just to be clear I'm understanding correctly: This is pulling the content of the RSS feeds of several news sites into the context window of an LLM and then asking it to summarize news items into articles and fill in the blanks? I'm asking because that is what it looks like, but AI / LLMs are not specifically mentioned in this blog post, they just say news are 'generated' under the 'News in your language' heading, whi…

Just for concrete confirmation that LLM(s) are being used, there's an open issue on the GitHub repository, on hallucinations with made up information, where a Kagi employee specifically mentions "an LLM hallucination problem":

https://github.com/kagisearch/kite-public/issues/97

There's also a line at the bottom of the about page at https://kite.kagi.com/about that says "Summaries may contain errors. Please verify important information."

Re: Kagi News

#212

Post by notsydonia » Tue, Sep 30, 2025, 6:24 PM UTC

rsingel wrote: post #165

Looks like yet another aggregator that will send almost no referral traffic to the folks who actually put labor into writing. Parasitic by definition. And embracing the news from nowhere perspective. So both a parasite and boring at the same time. I wish more tech folks who want to "fix the news" would learn from Gabe Rivera's Techmeme, Memeorandum, and Mediagazer. He's done aggregation right for 20 years

Exactly. I also wonder what the end game is. If creating content becomes a loss-making exercise, people will logically stop and the LLMs will have less and less to content to 'train on.' And as even large news corps are increasingly deploying internal LLMs, the deadening banal style of LLMs, A.I. over-view etc will inevitably drive readers away. I use Perplexity for search in place of Google and it surfaces good links most of the time. But what do tech and media companies - even spotify - think they will do when the artists, reporters and creatives stop feeding them? Or readers don't want to read banal summaries of everything?

Re: Kagi News

#213

Post by rdlw » Tue, Sep 30, 2025, 6:26 PM UTC

cogman10 wrote: post #86

__jonas wrote: post #72

Earlier quoted context omitted.

I have yeah, to me it looks like what I described in my comment above, it's LLM generated text, is it not? > And pretty much every single sentence cites the original article(s). Yeah but again, correct me if I'm wrong, but I don't think asking an LLM to provide a source / citation yields any guarantee that the text it generates alongside it is accurate. I also see a lot of text without any citations at all, here are…

Oof, that one is particularly bad because it cites 3 sources which are all the same article. Google points to phys and phys is a republish of the MIT article.

There's even a 'perspectives' section that tries to contrast two of the sources, but they're the same article.

Re: Kagi News

#214

Post by input_sh » Tue, Sep 30, 2025, 6:30 PM UTC

amarant wrote: post #161

devmor wrote: post #37

Earlier quoted context omitted.

The line “news stories will be generated” throws up red flags across the horizon for me. That’s not news. That’s news-adjacent random slop.

And yet, after trying it, I have to admit it's more informative and less provocative than any other news source I've seen since at least 2005. I don't know how they do it, and I'm not sure I care, the result is they've eliminated both clickbait and ragebait, and the news are indeed better off for it!

Soulless, uncreative, not fact-checked (or read by anyone before clicking publish), not contributing anything back to the original journalists, all of the editorial decisions are done by an undeterministic AI filter.

Not gonna call it the worst insult to journalism I've ever seen because I've seen factually(.)so which does essentially the same thing but calls it an "AI fact check", but it's not much better.

It's like instead of borrowing a book from the library, there's like a spokesperson at the entrance who you ask a question and then blindly believe whatever they say.

Re: Kagi News

#215

Post by rdlw » Tue, Sep 30, 2025, 6:31 PM UTC

atonse wrote: post #105

__jonas wrote: post #72

Earlier quoted context omitted.

I have yeah, to me it looks like what I described in my comment above, it's LLM generated text, is it not? > And pretty much every single sentence cites the original article(s). Yeah but again, correct me if I'm wrong, but I don't think asking an LLM to provide a source / citation yields any guarantee that the text it generates alongside it is accurate. I also see a lot of text without any citations at all, here are…

They are LLM generated summaries, sure. I guess I'm trying to understand your comment. Is there a distinction you're making between LLM summaries or LLM generated text, or are you stating that they aren't being transparent about the summaries being generated by LLMs (as opposed to what? human editors?). Because at some point when I launched the app, it did say summaries might be inaccurate. Looks like you found an ex…

> as opposed to what? human editors?

...yes? If I go to a website called "_ News" (present company included), I expect to see either news stories aggregated by humans or news stories written and fact checked by humans. That's why newspapers have fact checking departments, but they're being replaced by something with almost none of the utility and its proponents are framing the benefits of the old system as impossible or impractical.

Re: Kagi News

#216

Post by amarant » Tue, Sep 30, 2025, 6:31 PM UTC

I really like this!

I do wish I could have better control of what languages I'm getting. Right now the option is to either translate everything or nothing. I'd prefer news in their original, untranslated form if it's one of the 4 languages I speak, otherwise translate them to English.

I added the category "Israel" and everything was in Hebrew, so I had to set my language to English, but now news in my native Swedish are translated to English and I have to kind of translate it back in my head as I read them.

It's not the end of the world, but it seems like fairly low-hanging fruit!

Re: Kagi News

#217

Post by guywithahat » Tue, Sep 30, 2025, 6:32 PM UTC

One thing I found working on a startup which touched on the political sphere is people don't want curated lists imposed on them, they want to impose their curated lists on other people.

I like that it only provides the list once a day (I do think that's a clever feature), but the inability to influence bias seems like a mistake, especially since the sources already seem to follow a bias.

Re: Kagi News

#218

Post by carlosjobim » Tue, Sep 30, 2025, 6:33 PM UTC

"Community-driven sources: Our news sources are open source and community-curated through our public GitHub repository. Anyone can propose additions, flag problems, or suggest improvements."

This sounds like it's going to be a massive headache. Activists with nothing to do all day will be all over this, for their chance to try to have influence over what other people read.

Re: Kagi News

#219

Post by k2enemy » Tue, Sep 30, 2025, 6:34 PM UTC

I like 1440 (https://join1440.com) for this. Once a day daily email digest. I like the email format because I'm less likely to start clicking around compared to a web site, and it doesn't require a separate app.

I think it is human curated, but I'm not positive about that.

Re: Kagi News

#220

Post by edaemon » Tue, Sep 30, 2025, 6:35 PM UTC

stavros wrote: post #202

lukeschlather wrote: post #196

Earlier quoted context omitted.

They don't explicitly say they generate summaries at any point in the article. In fact I read it and though this was just some fancy RSS aggregator. The way they describe the "daily briefing" is extremely ambiguous.

OK, but I'd like to repeat my question here: Why do you care how the summary was generated?

I'm not the person you asked, but it's useful to know if the summary was generated using a method prone to inaccuracy.