3 ms·

I'm genuinely asking, but have you tried it? https://kite.kagi.com https://kite.kagi.com It actually seems more like an aggregator (like ground.news) to me. An

by atonse 1y ago

I'm genuinely asking, but have you tried it? https://kite.kagi.com https://kite.kagi.com It actually seems more like an aggregator (like ground.news) to me. And pretty much every single sentence cites the original article(s). There are nice summaries within an article. I think what they mean is that they generate a meta-article after combining the rest of them. There's nothing novel here. But the presentation of the meta-article and publishing once a day feel like great features.

__jonas 1y ago
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 three sections (Historical background, Technical details and Scientific significance) that don't cite any sources: https://kite.kagi.com/s/5e6qq2 https://kite.kagi.com/s/5e6qq2
- cogman10 1y ago
  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.
  - rdlw 1y ago
    There's even a 'perspectives' section that tries to contrast two of the sources, but they're the same article.
    - cogman10 1y ago
      I can envision the day where an LLM article generator starts consuming LLM generated articles which were sourced from single articles (co-written by an LLM).
- atonse 1y ago
  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 example where it isn't properly citing the summaries. My guess is that they will tighten this up, because I looked mostly at the first and second page and most of those articles seemed to have citations in the summaries. Like most people, I would want those everywhere to guard against potential hallucinations. 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. But another sibling commenter talked about the phys.org and google both pointing to the same thing. I agree, and this is exactly an issue I have with other aggregators like Ground.news. They need to build some sort of graph that distills down duplicates. Like I don't need the article to say "30 sources" when 26 of them are just reprints of an AP/Reuters wire story. That shouldn't count as 30 sources.
  - __jonas 1y ago
    > 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?). The main point of my original comment was that I wanted to understand what this is, how it works and whether I can trust the information on there, because it wasn't completely clear to me. I'm not super up to date with AI stuff, but my working knowledge is that I should never trust the output of an LLM and always verify it myself, so therefore I was wondering if this is just LLM output or if there is some human review process, or a mechanism related to the citation functions that makes it output of a different, more trusted category. I did catch the message on the loading screen as well now, I do still think it could be a little more clear on the individual articles about it being LLM generated text, apart from that I think I understand somewhat better what it is now.
  - jtbayly 1y ago
    > 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.
hadlock 1y ago
Publishing once a day to remove the "slot machine dopamine hit" is worth it for that alone. I have forever been looking for a peer/replacement to Google News, I was about to pony up for a Ground News subscription but I'll probably hold off for a couple more months. Alternatives to google news have been sorely lacking for over a decade, especially since google news got their mobile-first redesign which significantly and permanently weakened the product to meet some product manager's bonus-linked KPI. One more product to wean off the google mothership. Gmail is gonna be real hard though.
- jacobgkau 1y ago
  > Gmail is gonna be real hard though. Gmail seems like the easiest piece of the Google puzzle to replace. Different calendar systems have different quirks around repeating events, you sometimes need to try a variety of search engines to find what you're looking for, Docs aren't bug-for-bug equivalent to the Office or iCloud competitors, YouTube has audience, monetization, and hosting scale... Gmail is just "make an email account with a different provider and switch all of your accounts to use the new address." They don't even give you that much storage for free Gmail; it's 15GB, which lots of other email providers can match (especially paid ones). You can import your old emails to your new provider or just store them offline with a variety of email clients. Is updating all of your accounts (and telling your contacts about the new address) what you consider to be the hard part, or do you actually use any Gmail-specific features? Genuinely curious, as I tend to disregard almost all mail-provider-specific features that any of my mail providers try to get me excited about (Gmail occasionally adds some new trick, but Zoho Mail is especially bad about making me roll my eyes with their new feature notifications).
  - wintermutestwin 1y ago
    I am sticking with this reprehensible company for email because their spam detection is awesome and I have found no clear measurements of detection to reasonably compare. I’d love to be proven wrong!
    - gnud 1y ago

by atonse 1y ago

__jonas 1y ago

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 three sections (Historical background, Technical details and Scientific significance) that don't cite any sources: https://kite.kagi.com/s/5e6qq2 https://kite.kagi.com/s/5e6qq2

cogman10 1y ago
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.
- rdlw 1y ago
  There's even a 'perspectives' section that tries to contrast two of the sources, but they're the same article.
  - cogman10 1y ago
    I can envision the day where an LLM article generator starts consuming LLM generated articles which were sourced from single articles (co-written by an LLM).
atonse 1y ago
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 example where it isn't properly citing the summaries. My guess is that they will tighten this up, because I looked mostly at the first and second page and most of those articles seemed to have citations in the summaries. Like most people, I would want those everywhere to guard against potential hallucinations. 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. But another sibling commenter talked about the phys.org and google both pointing to the same thing. I agree, and this is exactly an issue I have with other aggregators like Ground.news. They need to build some sort of graph that distills down duplicates. Like I don't need the article to say "30 sources" when 26 of them are just reprints of an AP/Reuters wire story. That shouldn't count as 30 sources.
- __jonas 1y ago
  > 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?). The main point of my original comment was that I wanted to understand what this is, how it works and whether I can trust the information on there, because it wasn't completely clear to me. I'm not super up to date with AI stuff, but my working knowledge is that I should never trust the output of an LLM and always verify it myself, so therefore I was wondering if this is just LLM output or if there is some human review process, or a mechanism related to the citation functions that makes it output of a different, more trusted category. I did catch the message on the loading screen as well now, I do still think it could be a little more clear on the individual articles about it being LLM generated text, apart from that I think I understand somewhat better what it is now.
- jtbayly 1y ago
  > 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.

hadlock 1y ago

Publishing once a day to remove the "slot machine dopamine hit" is worth it for that alone. I have forever been looking for a peer/replacement to Google News, I was about to pony up for a Ground News subscription but I'll probably hold off for a couple more months. Alternatives to google news have been sorely lacking for over a decade, especially since google news got their mobile-first redesign which significantly and permanently weakened the product to meet some product manager's bonus-linked KPI. One more product to wean off the google mothership. Gmail is gonna be real hard though.

jacobgkau 1y ago
> Gmail is gonna be real hard though. Gmail seems like the easiest piece of the Google puzzle to replace. Different calendar systems have different quirks around repeating events, you sometimes need to try a variety of search engines to find what you're looking for, Docs aren't bug-for-bug equivalent to the Office or iCloud competitors, YouTube has audience, monetization, and hosting scale... Gmail is just "make an email account with a different provider and switch all of your accounts to use the new address." They don't even give you that much storage for free Gmail; it's 15GB, which lots of other email providers can match (especially paid ones). You can import your old emails to your new provider or just store them offline with a variety of email clients. Is updating all of your accounts (and telling your contacts about the new address) what you consider to be the hard part, or do you actually use any Gmail-specific features? Genuinely curious, as I tend to disregard almost all mail-provider-specific features that any of my mail providers try to get me excited about (Gmail occasionally adds some new trick, but Zoho Mail is especially bad about making me roll my eyes with their new feature notifications).
- wintermutestwin 1y ago
  I am sticking with this reprehensible company for email because their spam detection is awesome and I have found no clear measurements of detection to reasonably compare. I’d love to be proven wrong!
  - gnud 1y ago