Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Kagi News

blog.kagi.com

Post reply on HN
Print view View original

181–190 of 454 posts

Re: Kagi News

#181

Post by b112 »

meowface wrote: post #102
mvieira38 wrote: post #47

Earlier quoted context omitted.

Yes, that's what it is. Kagi as a brand is LLM-optimist, so you may be fundamentally at odds with them here... If it lessens the issue for you, the sources of each item are cited properly in every example I tried, so maybe you could treat it as a fancy link aggregator

I consider myself a major LLM optimist in many ways, but if I'm receiving a once per day curated news aggregation feed I feel I'd want a human eye. I guess an LLM in theory might have less of the biases found in humans, but you're trading one kind of bias for another.

Don't worry, all those news articles are of course human curated.

(I say this sarcastically and unhappily)

Re: Kagi News

#182

Post by jacobgkau »

CaptainOfCoit wrote: post #152
whatamidoingyo wrote: post #123

Earlier quoted context omitted.

> when you ask an LLM to point to 'sources' for the information it outputs, as far as I know there is no guarantee that those are correct A lot of times when I ask for a source, I get broken links. I'm not sure if the links existed at one point, or if the LLM is just hallucinating where it thinks a link should exist. CDN libraries, for example. Or sources to specific laws.

> A lot of times when I ask for a source, They'll do pretty much everything you ask of them, so unless the text actually come from some source (via tool calls, injecting content into the context or other way), they'll make up a source rather than doing nothing, unless prompted otherwise.

Maybe they should be trained on the understanding that making up a source is not "doing what you ask of them" when you ask for a source. It's actually the exact opposite of the "doing what you asked, not what you wanted" trope-- it's providing something it thinks you want instead of providing what you asked for (or being honest/erroring out that it can't).

Re: Kagi News

#183

Post by hughes »

I don't understand how this is 100% free, no subscriptions, no purchase, apparently no ads or tracking and yet I'm also the "customer" and not the product. What's the catch?

Re: Kagi News

#184

Post by kkukshtel »

One of the best news sites (still running) that I use frequently is http://68k.news/ - it's sort of like this minus the AI summary and info part of the article.

It's just plain text web 1.0 page that uses some ranking algo to figure out the top stores of a given day across categories, and shows that headline and under it similar headlines across different news sources.

It used to pull in RSS from the sources so you could also read the articles in plaintext, but that broke a bit ago and the dev hasn't fixed it.

Regardless, I still find it a great site to quickly get up to speed on top stories of the day!

But also I really like (and pay for!) Kagi so happily support their own effort here.

Re: Kagi News

#187

Post by jtbayly »

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…

> No, the citations don't guarantee that there weren't any hallucinations, but if you read something that makes you go "huh" – the citations give you a low-friction opportunity to read more.

Either you mean every time you read something interesting (“huh”) you should check it. But in that case, why bother with reading the AI summary in the first place…

Or you mean that any time you read something that sounds wrong, you should check it. But in that case, everything false in the summaries that happens to sound true to you will be confirmed in your mind without you ever checking it.

Re: Kagi News

#188

Post by alphazard »

What everyone gets wrong about news curation is thinking people want the same news as everyone else, or "both sides" of a situation, or whatever mechanism for exposing them to things that someone else thinks are true.

What I actually want is a curated set of things that are useful to me personally given my situation. The most important things about my situation to give me useful news are things like: net worth, income, citizenship, family situation, where I live, what industries I work in, current investments, travel destinations, regulatory and political risks associated with any of those things, etc.

Because those are the things that dictate how the parts of the world I can't control are going to affect me (especially if I don't react). I don't want to hear about random things that aren't going to affect me when I'm looking at the news. Sometimes I want to learn new random/useless things for fun, but that's a leisure activity. It's totally separate from the "news", which is a thing that adults consume as a chore to better plan their lives.

The fundamental problem is that myself and others are not going to willing give out the personal information required to curate useful news feeds, so the news will always be filled with noise. Maybe local AI can help with that.

Re: Kagi News

#189

Post by b112 »

krainboltgreene wrote: post #110
BeetleB wrote: post #51

Earlier quoted context omitted.

It's not binary - it's a continuum. When you go to Google News, the way they group together stories is AI (pre-LLM technology). Kagi is merely taking it one step further. I agree with your concern. I see this as a convenient grouping, and if any interests me I can skip reading the LLM summary and just click on the sources they provide (making it similar to Google News).

It cannot be "one step further", because there's a clear break in reality between what Google News provides and Kagi provides. Google News links to an article that exists in our world, 100%, no chance involved. Kagi uses an LLM generate text and thus is entirely up to chance.

Devil's advocate here.

Do you know that's what they're doing? They are a search engine after all. They do run their own indexer, as well as cache results from other sources.

If they're feeding urls to an AI, why can't they validate AI output urls are real? Maybe they do.

Re: Kagi News

#190

Post by input_sh »

atonse wrote: post #109
input_sh wrote: post #96

Earlier quoted context omitted.

It's useful for the users, but tragically bad for anyone involved with journalism. Not that they're not used to getting fucked by search engines at this point, be it via AMP, instant answers, or AI overviews. Not that the userbase of 50k is big enough to matter right now, but still...

All this is doing is aggregating RSS feeds and linking to the original articles. So this might result in lower traffic for "anyone involved in journalism" – but the constant doomscrolling is worse for society. So I think we can all agree that the industry needs to veer towards less quantity and more quality.

RSS feeds are meant to be used by actual users, not regurgitated publicly. RSS readers at the very least have have author info visible and its users tend to be reported to website's analytics with a special user agent.
Post reply on HN