Rendered at 05:34:59 GMT+0000 (Coordinated Universal Time) with Wasmer Edge.
andai 7 hours ago [-]
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.)
yunusabd 5 hours ago [-]
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.
tusharnaik 15 minutes ago [-]
> Yeah, the YouTube algorithm kind of does this, but you unfortunately have no real control over it.
Wouldn't that argument hold true even if it was implemented here?
a13n 6 hours ago [-]
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?
andai 6 hours ago [-]
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.
conductr 2 hours ago [-]
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.
card_zero 36 minutes ago [-]
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.
waldothedog 1 hours ago [-]
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.
progmetaldev 6 hours ago [-]
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.
andai 5 hours ago [-]
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.
pests 4 hours ago [-]
> Yeah, the YouTube algorithm kind of does this, but you unfortunately have no real control over it.)
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.
Fnoord 4 hours ago [-]
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).
> as I get older, I find that I have less energy for
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.
nilslice 5 minutes ago [-]
cool. and, a neat side effect of catching this late is that the HN Update broadcast I heard (labeled "10/20/2024, 9:59:47 PM", assuming Pacific time) included & described itself in the audio. at the time of writing, it's a top story currently on the front page!
hammock 3 hours ago [-]
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
laidoffamazon 2 hours ago [-]
Im reminded of the talking heads in Batman Beyond that were presumably AIs.
noman-land 3 hours ago [-]
I hate it but I also love it.
tristenharr 1 hours ago [-]
Would be cool to create embeddings for historical HN posts, and then use a users favorite posts to personalize the post selection by averaging the embedding vectors for a users favorite posts then doing a cosine similarity search to select stories most likely to be of interest to a user.
Although it would be even better to use a users like history, but I’m not sure if/how those can be accessed.
Speaking of, I’m curious how other folks use embeddings. I know you can average multiple embeddings together, but is anyone else doing other translations and having success? Thinking of King - Man + Women = Queen, It seems a lot of the time I see questions being directly used as inputs for semantic search/RAG. I wonder if it might make sense to create a large set of question-answer pairs and embed them and then determine the average translation to move from “question space” to “answer space”, then when you embed questions you apply the translation on the embedding to move it into “answer space” before performing RAG, or maybe this would just add too much noise?
vunderba 21 minutes ago [-]
hmmm, I can't speak to people using word2vec in conjunction with RAG, but the other use case is actually pretty common. (you don't need to generate answers though in my experience).
For each document intended for ingestion into a vector database:
- Use an LLM to generate a list of possible questions that the document is capable of answering (essentially equivalent to generating a quiz)
- Map these question embeddings back to the original documents
- Store document, document chunks, question 1, question 2, etc. into the vector database
So now when a person queries your RAG, you have the direct link from user query -> doc chunks, but additionally the transitionary link from user query -> similar query -> doc chunk.
throwaway0665 3 hours ago [-]
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 is so neat. I been wanting to build something for HN and can't believe I didn't think of this. It was also cool too hear the update with a mention of HN Update as well considering it's treanding! :)
ag8 2 hours ago [-]
I started listening to it after this very submission became #1 on HN, so it was very meta to listen to it talk about how it might be making stuff up...
Great project!
olup 6 hours ago [-]
I did https://radio-hn.pages.dev/ last year in the same idea, just once per day. Posted it somewhere here at the time.
yunusabd 6 hours ago [-]
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!
Enjoyed this a lot! Particularly enjoyed the meta update about this thread. There are some neat suggestions here, but if this could come up on my podcast feed as-is I would listen to it.
yunusabd 5 hours ago [-]
Did anyone notice it reporting about itself now? This will definitely go into the archive :)
liamYC 1 hours ago [-]
Haha yes!!
n2d4 2 hours ago [-]
Hah! Here's what it had to say about itself this hour:
> In a more self-referential turn, we examine the launch of HN Update itself. This initiative aims to provide hourly news broadcasts, summarizing top stories from Hacker News. While listeners appreciate the concept of curated news, there are concerns about the accuracy of the summaries and the potential for bias in representation. Community discussion suggests that while the value of such a service is evident, trust in the accuracy of the content remains a critical issue.
kcrwfrd_ 24 minutes ago [-]
> in a more self-referential turn
:o wow, the AI is becoming self-aware!
wg0 4 hours ago [-]
Unrelated but reminds me of GTA V's radio channels in cars. The news talking about what you just did.
Would be great to have a playback speed button as well. (I can't sit through any audio at 1x.)
yunusabd 6 hours 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.
yunusabd 6 hours ago [-]
I added a basic speed setting, hope 3x is enough ;)
pugio 5 hours ago [-]
Excellent, thank you. This is something I can listen to!
seungwoolee518 21 hours ago [-]
Wow... Really great.
Can you add some historical archive to listen n hours ago (or morning briefing like) summary as a feature?
yunusabd 20 hours ago [-]
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.
tkgally 8 hours ago [-]
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.
Because of that, catching up on several days will also be unique stories.
yunusabd 6 hours ago [-]
That's a real useful resource. Thanks!
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.
seungwoolee518 18 hours ago [-]
Yeah. You're right.
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.
yunusabd 18 hours ago [-]
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!
vulcan01 6 hours ago [-]
If you do this, it might be interesting to publish this as a podcast. Hourly is too frequent for a podcast, but daily would be pretty good.
(Podcasts are just RSS feeds.)
yunusabd 5 hours ago [-]
Sounds good, I'll get back to you about that!
nurettin 20 minutes ago [-]
I wonder what it will do when it encounters a (year) postfix.
frays 4 hours ago [-]
Great idea! I've always wanted to catch up with tech news while walking/driving, particularly from HN.
personjerry 5 hours ago [-]
"First up a webgpu wifi simulator has emerged, although details remain sparse due to trouble extracting data from the linked resource"
Sounds about right
yunusabd 4 hours ago [-]
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.
srik 7 hours ago [-]
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.
yunusabd 4 hours ago [-]
That seems to be the general consensus, I think it makes sense to move in that direction and beef it up a little.
cchance 5 hours ago [-]
Next step heygen avatar's doing desk presentations with small videos overlayed to show the websites etc.
mirrorchair01 8 hours ago [-]
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?
yunusabd 7 hours ago [-]
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.
ewalk153 6 hours ago [-]
Did you add the “hi mom” bit at the end, or was that organic?
yunusabd 5 hours ago [-]
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.
jonahx 5 hours ago [-]
The AI voice is good. What are you using to generate it?
yunusabd 4 hours ago [-]
It's openAI's voice generator.
fullstackchris 7 minutes ago [-]
Not affilliated, but if you want a large selection of _really_ good text to speech voices, check out eleven labs. Only problem is it's not cheap :/
8organicbits 5 hours ago [-]
"although details remain sparse due to issues extracting content from the linked resource"
Unfortunate that it adds stuff like this, which doesn't seem helpful to the listener.
yunusabd 5 hours ago [-]
Yep, some pages can't be read for some reason or another. I already added support for PDFs, I guess I'll handle more error cases as they pop up.
the_arun 4 hours ago [-]
Fantastic idea! I really liked extracting meta information from Comments as well. Good luck!
syntaxing 4 hours ago [-]
This is awesome, I wish it did a rolling 8 hour instead that is updated once a hour if that makes sense.
yunusabd 4 hours ago [-]
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.
pan69 10 hours ago [-]
Nice. Would it be possible to add a link to the actual submission as it is being discussed?
yunusabd 9 hours ago [-]
Definitely! I was thinking to just show a list of all the links under the player, would that work for you?
shannifin 7 hours ago [-]
Was thinking the same thing. Very fun idea, nice work!
cylinder714 3 hours ago [-]
Wow—this is indistinguishable from an NPR piece.
synthoidzeta 8 hours ago [-]
Really neat! How was the waveform visualization achieved?
Yep! First time using it, seems really solid and does the job.
graycat 1 hours ago [-]
Sounds good. Listened to some of it -- good.
Suggestion, have better audio:
(1) Enunciate the words more clearly.
(2) To help with (1), slow down and speak fewer words per minute.
(3) At the end of a sentence, don't drop voice volume and enunciation clarity.
kelsey98765431 5 hours ago [-]
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.
yunusabd 4 hours ago [-]
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 [...]"?
breck 4 hours ago [-]
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.
Thanks for the video, it's really interesting to see a user test like this!
neonsunset 3 hours ago [-]
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.
knowaveragejoe 7 hours ago [-]
I like this idea a lot. Archived entries would be nice as others have suggested. It would also be nice to be able to control playback speed.
yunusabd 5 hours ago [-]
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.
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.)
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.
Wouldn't that argument hold true even if it was implemented here?
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 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.
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...
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.
Although it would be even better to use a users like history, but I’m not sure if/how those can be accessed.
Speaking of, I’m curious how other folks use embeddings. I know you can average multiple embeddings together, but is anyone else doing other translations and having success? Thinking of King - Man + Women = Queen, It seems a lot of the time I see questions being directly used as inputs for semantic search/RAG. I wonder if it might make sense to create a large set of question-answer pairs and embed them and then determine the average translation to move from “question space” to “answer space”, then when you embed questions you apply the translation on the embedding to move it into “answer space” before performing RAG, or maybe this would just add too much noise?
For each document intended for ingestion into a vector database:
- Use an LLM to generate a list of possible questions that the document is capable of answering (essentially equivalent to generating a quiz)
- Map these question embeddings back to the original documents
- Store document, document chunks, question 1, question 2, etc. into the vector database
So now when a person queries your RAG, you have the direct link from user query -> doc chunks, but additionally the transitionary link from user query -> similar query -> doc chunk.
1: https://news.ycombinator.com/item?id=41899828
Great project!
> In a more self-referential turn, we examine the launch of HN Update itself. This initiative aims to provide hourly news broadcasts, summarizing top stories from Hacker News. While listeners appreciate the concept of curated news, there are concerns about the accuracy of the summaries and the potential for bias in representation. Community discussion suggests that while the value of such a service is evident, trust in the accuracy of the content remains a critical issue.
:o wow, the AI is becoming self-aware!
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.
Can 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.
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.
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.)
Sounds about right
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.
Unfortunate that it adds stuff like this, which doesn't seem helpful to the listener.
Suggestion, have better audio:
(1) Enunciate the words more clearly.
(2) To help with (1), slow down and speak fewer words per minute.
(3) At the end of a sentence, don't drop voice volume and enunciation clarity.
> 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 [...]"?
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...