Thanks! An archive shouldn't be too hard to implement.
Would a morning briefing include more of a curated selection of news, like the most interesting/most discussed news of the day? Currently I just take the top 5 posts from the main page.
I was actually listening to it while making breakfast as well, so the idea of an extended morning briefing resonates with me.
I'll look into it, thanks for the follow-up!
I could easily see listening to such a summary once or twice a day. Considering the pace at which HN is updated, how about two to four somewhat longer episodes per day, with an archive of the past week so that people can catch up? You might also want to focus on certain types of stories, such as those with more upvotes or comments.
In any case, it's a great idea, and I enjoyed listening to the current episode.
This is really great stuff, love the added "news-station" type music! If something like this we're to be monetized would there be any issues regarding copyright?
Thank you! And that's a really good question. Since it's summarizing the articles, I would assume that there shouldn't be any issues regarding copyright. Regarding the comments that it's using, I think HN generally has some rights to them, although again it's more of a summarization. Generally HN seems cool about these things.
Now, if you were to scan the homepage of the BBC and create a news broadcast from it, they probably wouldn't be too happy about that, understandably. I have no idea how this would fare in court though.
I added it to the prompt for this exact situation. So it was "hard-coded". Tbh it would have been a bit scary if that happened organically. Kind of like animals recognizing themselves in the mirror levels of self-awareness.
Brilliant idea. I think this has real value as well: as I get older, I find that I have less energy for reading, but I also notice I often miss really cool stuff that was briefly on the front of HN.
I like that it summarizes the comments too. There are often real gems buried in there. (I assume you're only taking a few top ones?)
I think a great improvement could be made with personalization. Most of what's on the front page isn't personally relevant to me, and there's a lot of cool stuff on the new submissions page that never catches on. So it would be nice if a system could learn what kind of stories I personally respond to, and show me (a summary of?) those -- even if they aren't currently trending.
Last.fm came out 20 years ago (proving you don't even need AI for amazing recommender systems), but it seems personalizing your experience never really caught on. (Yeah, the YouTube algorithm kind of does this, but you unfortunately have no real control over it.)
agreed on energy for reading. do you think it’s that we’re getting older or that the friction associated with consuming information is just getting lower and lower over time?
So my energy levels have declined noticeably from age 20 to 30. I thought it was mostly my own chronic health issues causing accelerated aging, but many of my friends are making similar complaints.
I did notice far before this point (e.g. age 10 to 20) that my patience for reading had gone down significantly. If I had to guess I'd say that in my case it's due to an underlying anxiety that started in early teens and never left me. Drowning it out seems to require something more stimulating than reading (on paper).
(Perhaps meditation or therapy (shadow integration?) would help here. I've certainly had glimpses of inner peace during times when I was meditating regularly. One insight from this time: "holy crap, I always thought I needed to struggle harder, but it turns out I just needed to learn how to relax...")
I can get through audiobooks but I can only consume them while traveling or doing chores. If I'm sitting down, the restlessness is too high and I can't concentrate on the book.
I also have ADHD and heard similar things from others with ADHD, so I'm not sure to what degree this translates to others.
I never had ADHD or related attention problems. However, I find the way interwebs has evolved over years and to basically trained our brains to be so focused on super short pieces of simply digestible content has basically turned me off of longer form content. I’ve noticed my preference for immediate gratification has increased significantly. It feels like I’m developing an attention disorder in my middle age years.
I read a huge amount every day, lots of news articles, chapters of whatever book, random material of interest, people's comments and questions. I rarely read anything "long-form", however, because it has a strong tendency to be a giant self-indulgent bloviating pile of shit. There are exceptions: I've read every story on damninteresting, because it's true to its name. (Hi Alan. Post something new.)
I've nearly finished Moby Dick, but I don't know why, it's dreadful, I wouldn't recommend it to anyone. I guess it's historically interesting.
Possibly you guys are not losing the will or energy to read, you're merely discovering that the things you think you ought to read are terrible. Try only reading what you like. So what if it's short? Maybe your preference is right.
I can't read long form content from a phone screen or computer monitor. I really don't like it. It might be the scrolling, or maybe the type of screen, or conditioning to expect instant gratification. I do, however, read 20-30 books a year, both paper and on an ereader.
I think it's super important to make reading a habit so one must identify what doesn't work for them and try something else. You lose so much if you don't read.
Yeah, reading on a screen is ass. What eReader do you use?
I got a Kobo recently and I don't like it much. It's much duller than my last one, which is either due to the color screen having less contrast or me misremembering.
Kobo's main selling point was that it's not botnet, but I couldn't even turn the thing on without making an online account...
I use a Kobo Clara HD. I think Claras are the second cheapest ones. No colour. I find the contrast about as good as a cheap paperback book. Not as good as good quality printing on acid-free paper, but that's OK.
I actually prefer lower contrast, it gives me less eye fatigue and I've specifically chosen low-contrast themes on my computer for as long as I can remember.
I think there is a way to not create the account if you really don't want to. I put KOReader on mine straight away and never use the built-in software. That also incidentally has an option to adjust the contrast.
Wanted to share that I also had intense anxiety/depression well into my thirties but was able to finally sort it out a few years ago.
I think your phrase "learn how to relax" is spot-on. I was chronically under-slept and chronically over-focused on school, university, career, side-hustle(s), etc.--without much great effect I might add. My brain was so wound up from a decade-plus of hyper-vigilance that I had actually forgotten how to relax.
Obviously, I am not you and n=1, but hopefully it is encouraging to know that you might not always need to drown it out.
When it comes to online reading, there are quite a few things that cause me fatigue that I don't feel I used to experience. Advertisements have been there for a long time, but often these are woven into the content, either as literal text placed in the article, or as visual ads that you need to scroll through to continue reading the article. Relying on different JavaScript and CSS techniques to "enhance" the user experience often cause me issues when I'm just trying to focus on reading. Those include overriding scrollbars, dynamic loading of content when the text is small enough to have been included in the page, and displaying some kind of alternative action when highlighting text. I'll often highlight text to keep track of where I'm reading, and some sites will pop up a dialog with share actions, or the ability to add annotations, etc. This is distracting and makes it more difficult to follow along with a longer article.
I'm 45, and got my start on BBS pre-internet, but I feel like if I find an article without the distractions I mentioned above, I actually have more energy to complete an article than I did in my 20's and 30's. Having access to the article without distractions helps me to focus, and when I focus, I tend to consume more content than I normally would. Most likely one of the reasons we're drawn to HN.
You've just made me realize why I usually avoid clicking the actual article link on HN. It's usually a very unpleasant experience, unless it's clear that it's a smaller website.
Perhaps a small help for you might be "reader mode" or "focus mode" or whatever your browser of choice calls it.
It's usually embedded in the url bar (probably a hotkey for it), and gives you only the text. A major step forward in not having to subconsciously ingest and then choose to ignore all the ads, related links, etc.
Thank you, I'm not sure why I always seem to forget that feature. Although I don't like the Google lock-in, on certain sites I can also visit the AMP version of the page, and have fewer ads while also getting the images.
Thanks for the comment, I'm really enjoying the discussion it has sparked.
Yes, I'm just taking the top comments, along with a few child comments, in order to not exceed the context window of the model.
Regarding personalization, there's definitely lots of potential. HN can be so random though, sometimes you find things that you didn't even know you needed (intellectually). I guess as with most recommender systems, it's about a balance between exploration and exploitation.
Maybe an MVP could filter for specific keywords and add those posts to the model input.
This is possibly an interesting extension [1]. I just generally don't follow recommendations. But my kids love it, and they use my account (Premium, else they get all kind of inappropriate ads (which is even illegal)), so I have to be careful. So the other day I wanted to look into what the other political side had to say about something. You know, as a matter of broadening my view to gain some understanding. Boy, did I regret, as I was getting sucked into some kind of conspiracy bubble. My wife asked me what on earth I watched. So I ended up trying to have YouTube profile me as little as possible (via settings). Didn't fully solve it, but it is much more clean now. And if I do watch something which I don't want to be remembered: private browsing mode. I do this for porn, but unfortunately they do profile my IP address (so I should use a VPN).
I just deleted my account and switched to an old account and the quality of my recommendations improved by an order of magnitude.
It's probably because I last used that one 10+ years ago when it seems things were a little more lighthearted (at least in my digital world). Going back to that recommendations page in current year was a truly magical experience.
Of course it didn't last though, within a few weeks algorithm was onto me, and went back to showing me the same stuff as on my old account.
> as I get older, I find that I have less energy for reading,
As I get older (50 now), I have 0 patience for anything else than reading; it's much faster than listening and rereading stuff I didn't get the first time is easier and faster. I listen to stuff (including zoom/calls) to stuff that doesn't hold much importance/value; maybe it triggers something, then I go read about it instead.
Actually, I used to love reading, but the last years I've felt exactly like I had no energy to read any more, and have switched to audiobooks and podcasts. I then checked my eyes, and it turns out I need glasses. So as you get older, check your eyes!
I have had glasses since I was 7... A year after my father brought home a 'portable computer' (luggable) [0] with a tiny monochrome screen. Not sure if that was related, but I guess it could be as the screen was very tiny (see pic below).
I think people have different definitions of "reading":
- Skimming: In my day job this is 95% of "reading". I think this is unavoidable, since (even if the material is well written) there's very little chance that two consumers need the same information. In papers I read the abstract, jump to the conclusions, maybe go back to the intro if I'm confused, or check out the methods if that's what matters to me. I get frustrated with any medium where skimming isn't possible, and similar when search isn't possible.
- Reading as in reading a book: this is more for fun or to cool down. Video / podcast seems like a drop-in replacement here. I don't read HN this way. Does anyone?
I think people who create content should be aware of this dichotomy. If you are communicating with experts, make sure your information is well structured. If you are writing literature focus more on the flow.
I skim HN, when I like something, I dive in & read it carefully and sometimes even implement it. The rest is just chewing gum; nice maybe but not serious.
> Last.fm came out 20 years ago (proving you don't even need AI for amazing recommender systems)
What makes that recommender system not AI? Wikipedia say it uses "collaborative filtering" which Gemini says is a form of AI. AI started nearly 70 years ago.
Same boat re: audio speed. I actually speed up the voice in the backend by 1.16x . Above that I was getting too many artifacts in the audio. The nice thing about doing it at that point is that I can handle the music and the voice separately, i.e. the speed of the music stays unchanged.
Speeding it up in the player will also speed up the music, which is not very zen. But I guess I'll just add it to the player and let people decide how fast they want to go.
I added a basic speed setting, have a look! Also looking into the archive idea, which will be a bit more involved, since my current infra is kind of minimalist.
Neat! Personally hourly feels a bit much but a daily briefing that can fit inside a commute or on a short walk would be perfect. Might be the first ai podcast I'd subscribe to.
Interesting, I actually did a search before submitting mine, but I narrowed it down to the last year only. Yours being 2y/o didn't show up. You were ahead of your time!
Damn, looks like you beat NotebookLM by a year and a half!
Is the code for this available? I'm particularly curious how you did the multiple speakers and voices.
NotebookLM has the issue that they keep switching sides, like one will be the student and the other the teacher on a subject, but then they'll suddenly switch in a way that makes no sense.
I did not release the code but it's incredibly basic, and I believe OP's one is just the same.
You collect N links from HN api with any heuristic you want, then scrape those urls - preferably using pupeteer-based tooling or online equivalent (think Jina).
I then ran each url's content in an LLM to get a summary, then from all the results ask a LLM to create the conversation (and give it a tone). Then decide on the voices and characters and feed each turn into 11labs (or any tts). And finally, concatenate all audio parts, add music and effects.
If I remember correctly, mine could perform all that from a single Cloudflare worker. The catch is it can become a bit pricey because of the TTS. I remember toying with making it a product (podcast everything) and quickly discovered there's a couple of company already offering this.
NotebookLM is slightly different on the TTS front, I think they are using the amazing model google showed off a year or so ago (without giving it public access) that can generate actual multi speakers conversations with "hums" and cutting, and talking at the same time.
Frankly this is super fantastic. Thank you. Any possibility to make this longer, and split it into sections with updates on previous stories at the end? I love this concept a lot.
Thank you! Most of the people here seem to prefer a longer form, so I think I'll move in that direction.
> split it into sections with updates on previous stories at the end
So basically, if a story comes up that has been reported on before, it references the older story and gives some context and the update? Something like "just last week, the Internet Archive had been the victim of another cyber-attack, in which [...]"?
Yeah, if you look at the page source, there's actually no text in the document body. At one point I will have to use a JS-capable browser, to capture pages like this one.
Fun stuff. It does feel like NotebookLM (and others) are hurtling us toward a future that seems inevitable: all content is public domain, and people consume it in many transformed ways.
Those that get ahead of the curve and make their content publicly available and semantically well structured will see their ideas thrive.
So the best of the last 8 hours, with a new episode every hour? I think that makes sense, you could check in every 8 hours or so and have a fresh set of stories.
This is so good! I could use that mixed in together with a spotify station for modern Radio experience. I miss the days where you could just do something else and listen to the radio host. Streaming partially supplanted it but not to the same extent, sadly.
Pretty neat but it seems to make stuff up. It took a meta comment from this post[1] about the website formatting and suggested the community was worried the C++ memory safety proposal would make code hard to read on mobile. It is hard to trust the other summaries after hearing that.
> This formatting and font mixing is difficult to read on mobile.
I guess for us it's obvious that it's a meta comment, but I can understand the confusion. Still, it could have figured out that "readability on mobile" doesn't really apply to C++, a programming language.
I'll add a section to the prompt reminding it that comments can be meta or even non-factual (gasp), so that it doesn't try to shoehorn meaning into comments like this one.
Apparently you can add "don't make stuff up" to the prompt and it helps. I'm not sure the exact phrasing but probably something like only using what's in the text given.
Next step- create an HNN news network with chyrons and AI news anchors that you can have running on the TVs in your office 24/7 the way banks have CNBC and Bloomberg News running
I would actually use that, especially if it had hourly summaries of the days news like some real news networks do, so you don’t have to watch all day but could just chuck it on when convenient.
The best part about these AI-reads-content things is the potential to let users do things like choose the voice, the talking speed, etc. As a non-American I hate listening to those American talking-head news channels because it’s jarring for me when I’m used to my own accent, even if I do want to catch up on some current events or something.
This strikes a balance with one's use of time and "fear of missing out" (FOMO). Thank you, yunusabd, for creating this service. As a side note, I've been following HN headlines and comments for a very long time, but only just now created an account, just so that I could write this comment to express my thanks to you. On the subject of wishing for additional features: I agree with someone else who commented about how it would be nice to have some form of a daily summary, or maybe broken into three portions of summaries per day: morning, afternoon and evening. Additionally, those three portions would also (ideally) remain available and not be removed, so that a person might go back and review a particular range of times/days. This would be much more practical to enable a person to be able to keep up with most of the developments, but without needing to check in once every hour.
184 comments
[ 2.3 ms ] story [ 202 ms ] threadCan you add some historical archive to listen n hours ago (or morning briefing like) summary as a feature?
Would a morning briefing include more of a curated selection of news, like the most interesting/most discussed news of the day? Currently I just take the top 5 posts from the main page.
My morning routine is Checking HN with Coffee. So with your service I can minimize some time to click around and figure out what the root cause is.
(Podcasts are just RSS feeds.)
In any case, it's a great idea, and I enjoyed listening to the current episode.
Because of that, catching up on several days will also be unique stories.
I was looking into using the algolia API to get the top 10 posts of the last day, but those won't be unique over the span of several days.
Now, if you were to scan the homepage of the BBC and create a news broadcast from it, they probably wouldn't be too happy about that, understandably. I have no idea how this would fare in court though.
I like that it summarizes the comments too. There are often real gems buried in there. (I assume you're only taking a few top ones?)
I think a great improvement could be made with personalization. Most of what's on the front page isn't personally relevant to me, and there's a lot of cool stuff on the new submissions page that never catches on. So it would be nice if a system could learn what kind of stories I personally respond to, and show me (a summary of?) those -- even if they aren't currently trending.
Last.fm came out 20 years ago (proving you don't even need AI for amazing recommender systems), but it seems personalizing your experience never really caught on. (Yeah, the YouTube algorithm kind of does this, but you unfortunately have no real control over it.)
I did notice far before this point (e.g. age 10 to 20) that my patience for reading had gone down significantly. If I had to guess I'd say that in my case it's due to an underlying anxiety that started in early teens and never left me. Drowning it out seems to require something more stimulating than reading (on paper).
(Perhaps meditation or therapy (shadow integration?) would help here. I've certainly had glimpses of inner peace during times when I was meditating regularly. One insight from this time: "holy crap, I always thought I needed to struggle harder, but it turns out I just needed to learn how to relax...")
I can get through audiobooks but I can only consume them while traveling or doing chores. If I'm sitting down, the restlessness is too high and I can't concentrate on the book.
I also have ADHD and heard similar things from others with ADHD, so I'm not sure to what degree this translates to others.
I've nearly finished Moby Dick, but I don't know why, it's dreadful, I wouldn't recommend it to anyone. I guess it's historically interesting.
Possibly you guys are not losing the will or energy to read, you're merely discovering that the things you think you ought to read are terrible. Try only reading what you like. So what if it's short? Maybe your preference is right.
I think it's super important to make reading a habit so one must identify what doesn't work for them and try something else. You lose so much if you don't read.
I got a Kobo recently and I don't like it much. It's much duller than my last one, which is either due to the color screen having less contrast or me misremembering.
Kobo's main selling point was that it's not botnet, but I couldn't even turn the thing on without making an online account...
I actually prefer lower contrast, it gives me less eye fatigue and I've specifically chosen low-contrast themes on my computer for as long as I can remember.
I think there is a way to not create the account if you really don't want to. I put KOReader on mine straight away and never use the built-in software. That also incidentally has an option to adjust the contrast.
I think your phrase "learn how to relax" is spot-on. I was chronically under-slept and chronically over-focused on school, university, career, side-hustle(s), etc.--without much great effect I might add. My brain was so wound up from a decade-plus of hyper-vigilance that I had actually forgotten how to relax.
Obviously, I am not you and n=1, but hopefully it is encouraging to know that you might not always need to drown it out.
I'm 45, and got my start on BBS pre-internet, but I feel like if I find an article without the distractions I mentioned above, I actually have more energy to complete an article than I did in my 20's and 30's. Having access to the article without distractions helps me to focus, and when I focus, I tend to consume more content than I normally would. Most likely one of the reasons we're drawn to HN.
It's usually embedded in the url bar (probably a hotkey for it), and gives you only the text. A major step forward in not having to subconsciously ingest and then choose to ignore all the ads, related links, etc.
If you're eating snacks all day, you won't have a healthy appetite for proper meals.
I noticed on vacation I spend far less time online, and a lot more time reading books.
Yes, I'm just taking the top comments, along with a few child comments, in order to not exceed the context window of the model.
Regarding personalization, there's definitely lots of potential. HN can be so random though, sometimes you find things that you didn't even know you needed (intellectually). I guess as with most recommender systems, it's about a balance between exploration and exploitation. Maybe an MVP could filter for specific keywords and add those posts to the model input.
You do, indirectly. Just need to curate your last watched videos.
Sometimes I feel like I got put into a certain genre or bubble or if things autopplay a when I sleep I'll not ice my front page being taken over.
I just go delete some of those videos from my recent list I can see visible improvement.
Or just start a few video on the topic you want to see and then it's all you'll be recommended.
[1] https://addons.mozilla.org/en-US/firefox/addon/regretsreport...
It's probably because I last used that one 10+ years ago when it seems things were a little more lighthearted (at least in my digital world). Going back to that recommendations page in current year was a truly magical experience.
Of course it didn't last though, within a few weeks algorithm was onto me, and went back to showing me the same stuff as on my old account.
Side comment: When a person says something like that, they might be speaking of only themself, but there's a different parsing that many will hear.
Ageism is a real problem in our field, and one thing we can do is to not accidentally feed it.
Wouldn't that argument hold true even if it was implemented here?
As I get older (50 now), I have 0 patience for anything else than reading; it's much faster than listening and rereading stuff I didn't get the first time is easier and faster. I listen to stuff (including zoom/calls) to stuff that doesn't hold much importance/value; maybe it triggers something, then I go read about it instead.
[0] https://en.wikipedia.org/wiki/Osborne_1
- Skimming: In my day job this is 95% of "reading". I think this is unavoidable, since (even if the material is well written) there's very little chance that two consumers need the same information. In papers I read the abstract, jump to the conclusions, maybe go back to the intro if I'm confused, or check out the methods if that's what matters to me. I get frustrated with any medium where skimming isn't possible, and similar when search isn't possible.
- Reading as in reading a book: this is more for fun or to cool down. Video / podcast seems like a drop-in replacement here. I don't read HN this way. Does anyone?
I think people who create content should be aware of this dichotomy. If you are communicating with experts, make sure your information is well structured. If you are writing literature focus more on the flow.
I skim HN, when I like something, I dive in & read it carefully and sometimes even implement it. The rest is just chewing gum; nice maybe but not serious.
What makes that recommender system not AI? Wikipedia say it uses "collaborative filtering" which Gemini says is a form of AI. AI started nearly 70 years ago.
Would be great to have a playback speed button as well. (I can't sit through any audio at 1x.)
Speeding it up in the player will also speed up the music, which is not very zen. But I guess I'll just add it to the player and let people decide how fast they want to go.
Is the code for this available? I'm particularly curious how you did the multiple speakers and voices.
NotebookLM has the issue that they keep switching sides, like one will be the student and the other the teacher on a subject, but then they'll suddenly switch in a way that makes no sense.
You collect N links from HN api with any heuristic you want, then scrape those urls - preferably using pupeteer-based tooling or online equivalent (think Jina).
I then ran each url's content in an LLM to get a summary, then from all the results ask a LLM to create the conversation (and give it a tone). Then decide on the voices and characters and feed each turn into 11labs (or any tts). And finally, concatenate all audio parts, add music and effects.
If I remember correctly, mine could perform all that from a single Cloudflare worker. The catch is it can become a bit pricey because of the TTS. I remember toying with making it a product (podcast everything) and quickly discovered there's a couple of company already offering this.
NotebookLM is slightly different on the TTS front, I think they are using the amazing model google showed off a year or so ago (without giving it public access) that can generate actual multi speakers conversations with "hums" and cutting, and talking at the same time.
Added : someone just made a python lib for the usecase (also found on HN 5 days ago) https://github.com/souzatharsis/podcastfy
> split it into sections with updates on previous stories at the end
So basically, if a story comes up that has been reported on before, it references the older story and gives some context and the update? Something like "just last week, the Internet Archive had been the victim of another cyber-attack, in which [...]"?
Unfortunate that it adds stuff like this, which doesn't seem helpful to the listener.
Sounds about right
Those that get ahead of the curve and make their content publicly available and semantically well structured will see their ideas thrive.
Here's my user test: https://news.pub/?try=https://www.loom.com/embed/9e8b8d454ee...
1: https://news.ycombinator.com/item?id=41899828
> This formatting and font mixing is difficult to read on mobile.
I guess for us it's obvious that it's a meta comment, but I can understand the confusion. Still, it could have figured out that "readability on mobile" doesn't really apply to C++, a programming language.
I'll add a section to the prompt reminding it that comments can be meta or even non-factual (gasp), so that it doesn't try to shoehorn meaning into comments like this one.
The best part about these AI-reads-content things is the potential to let users do things like choose the voice, the talking speed, etc. As a non-American I hate listening to those American talking-head news channels because it’s jarring for me when I’m used to my own accent, even if I do want to catch up on some current events or something.