Initially I asked ChatGPT to estimate three things: event scale, event magnitude and event potential. That often resulted in clickbait articles going to the top.
To fix this I started to also ask it to estimate source credibility, so tabloids would get much lower score than, say New York Times.
Now you noticed another problem, similar articles get very different scores. I think ideally I could do some sort of deduplication, but I don't know how to implement it yet.
Oh, sliders for custom scoring is an amazing idea. And should be easy to add — I already have all the ChatGPT estimations for different parts of the score. Added to the TODO list.
The problem with deduping is that some news get posted and reposted by different sources for several days (sometimes even weeks) in a row. That's a huge context I'd have to put in AI.
1000 news titles * 3 days * 70 symbols per title = 210,000 symbols = 40000 words = 53000 tokens.
My current context window is 8000 tokens and I think 32000 tokens is max that GPT-4 allows.
---
Add: now that I think about it should be possible to do in several runs. Will keep thinking about it, thanks for the suggestion.
Maybe it's my low prompting skills, but I couldn't make ChatGPT give lower importance to tabloid articles when they claimed that the "world is ending" or something similar.
Every time ChatGPT saw the words "World is ending" (not real example) it gave those articles very high score.
Estimating source credibility was the only solution I came up with.
You could deduplicate based on vector similarity within a few days (7 days for big stories as their news cycle is longer, 24 or 48 hours for smaller stories).
Idea: The number of stories within a cluster weighted by credibility of the source could be another element of the rating.
Agreed on the point, but your example brings up a different problem for me.
100,000 dead Russians, is very likely the propaganda number. Maybes it’s right; maybe it’s not, IDK, but I know that there has been zero negative Ukraine news since the conflict started (ghost of Kiev anyone?), and the media machine definitely only works one way on this topic.
I mean, Russia isn’t the good guy here, but I don’t excuse propaganda just because it tells a story I want to hear.
In this case, I don’t care what the US or Ukraine says the number is. That isn’t news. It’s narrative, true or not.
So, I feel like the score should take into account if how likely an article is to be narrative vs purely an event.
The estimate is 100 000 casualities (20k dead and 80k wounded), and it seems to be correlated with several different intelligence sources.
The number is news because Russia is waging a large-scale invasion of another country using cannon fodder tactics. Reporting on that invasion is not simply propaganda.
(below is the project description I used when posted about it on Reddit)
The problem I have with most news sites is that I can't read only important news: an article about a virus outbreak is followed by some celebrity gossip or another smartphone release.
But even on sites that focus on important events articles are posted every day and there are always "top headlines" — even on days when nothing important happened.
I am forced to make a choice: waste time going through unimportant updates or ignore the news and miss important events.
So I built a web app that I think solves this.
It uses AI (ChatGPT-4) to read the top 1000 news every day and rank them by significance on a scale from 0 to 10 based on event magnitude, scale, potential, and source credibility.
I also run a newsletter where I post summaries of all the news with a score over 6.5. On average that's 1-3 articles per day, but sometimes it is 5, and sometimes — none at all. In that case, I just send an email saying that nothing important happened that day.
Yes! I do exactly that. In the early version I posted the original titles, but many were too clickbait-y and hard to read. I find the rewritten titles much easier to scan through.
How do you decide how aggressively to summarize? That is, a 200-word summary of an 800-word article is very different from a 200-word summary of a 5,000-word article.
Relatedly, it would be neat if you had a slider or +/- buttons for each article summary, so someone could choose their depth based on how interesting the title sounds.
I just ask chatgpt to "summarize the article very concisely" and hope that it won't lose any important points. But chatgpt is really good at summarizing, so I don't worry much.
Having +/- buttons is a good idea, will add it to todo list as well.
This is what I was thinking, sort of like the mechanism that Newsela and others use for their leveled reading platforms.
If you were so inclined, it would probably be pretty trivial to also offer leveled versions as well, which would be very useful for schools/teachers to use with students. There are a handful of companies that have bespoke leveled reading articles created from standard news articles. ChatGPT would make the workflow for creating news articles much easier, though manual QC would still be necessary to ensure that the article summaries are accurate and appropriate for students.
A great use case for llms. I'm not expecting you to do this, but I would love to choose my own news sites. I'm a Brit living in Czech Republic. So I'd maybe have the guardian, BBC, and some Czech sources in there. Not so interested in American news, any American news that affects me will reach my local sources. But great work!
A prompt, or more like a list of them seems the most interesting customization. For example: "no crypto currency", "nothing from Keith Rupert Murdoch", "chinese realestate", "ukraine war", "forex" etc
Now that would be worth something...
Say, you want more than 6 filters they costs 50 cents/filter/month with a minimum of 5.
I think one would gradually make more and more of them?
This is an awesome idea and I'm kicking myself for not thinking about it before. Very creative.
Have you made any attempt at quantifying its biases? Both whether it considers left- or right-leaning articles more significant and where it focuses most of its reporting. (Economy and the US from a first glance?)
> Have you made any attempt at quantifying its biases?
I haven't. I can share ChatGPT's evaluation of sources credibility, maybe that will give people some insights.
For example, Reuters got 9.5 (out of 10), Lifehacker 7, Oprah Mag 3.
> where it focuses most of its reporting. (Economy and the US from a first glance?)
> Have you made any attempt at quantifying its biases? Both whether it considers left- or right-leaning articles
FWIW, when I read news I often cross check on different outlets to get a sense of the span, range and sometimes media blackouts, including the ones with extreme bias (like RT reporting on Ukraine). This can give a good reading of the meta-temperature of a developing or controversial situation with a lot of propaganda and bias. In some cases, less trustworthy outlets will cover stories that aren’t narrative-friendly to more reputable publications, and sometimes, those can be really important.
For a service like this, I’d much prefer something analogous to “here’s what different outlets are saying”, rather than trying to make its own judgment about bias.
Great idea, the “here’s what different outlets are saying” feature.
Stories that aren’t narrative-friendly to main stream media.
Media blackouts to highlight propaganda and bias.
Make it easy for reader to see the "manufactured consent"!
I actually created a site - https://detoxed.news - with a very similar philosophy a while back, though with a much simpler implementation. It periodically scrapes Wikipedia's Current Events Portal (https://en.wikipedia.org/wiki/Portal:Current_events) and presents that information in a nicer format.
I think there's definitely a sweet spot where you keep well-informed about current events and happenings in the world, without wasting your time on the 24 hours new cycle.
The problem with using Wikipedia as a news source is that it focuses on really flashy topics, like murder and armed conflicts. Looking at the headlines of detoxed.news confirms this:
1) Flood in Rwanda kills 109 people 2) Sudan War 3) Australia evacuation from Sudan 4) Seizing of an oil tanker
None of these headlines keep me particularly informed and are actually a more potent version of the 24 hours new cycle
I get the most random articles by selecting around the mode (peak) of the distribution. The lowest of the low is all Daily Mail gossip all the time :puke:.
very good idea and good website, but very US focused. so not very useful. one thing that people tend to ignore about google and apple is how good their localization are. ofc i am not expecting the same sophistication from a team of 1, and i am not expecting anything at all. but i just want to encourage developers to think global from the get go.
Some news entries appear to be of the "Top 5 things you should know today" variety, where it covers multiple topics. Example screenshot.[0] Might want to filter those out somehow?
Great work! I'd consider specifying the list of criteria for significance further, in order to counteract the heavy US bias of ChatGPT and, therefore, your site. I'm not a prompt engineer, but perhaps something on relevance for people from different countries, or simply adding to your prompt a sentence on ensuring geographical diversity.
The idea of filtering news sourced mentioned elsewhere is nice, but will necessitate considerable input from users, kind of negating the purpose of the site
I have a question: when watching shows like Last Week Tonight by John Oliver, it feels like the important news that make direct impact on people’s lives are more local than national or even international. The state changing a law here, the city building something new or make it harder to X..etc.,
Have you considered/thought about this aspect of news?
I bet spam algorithms didn't like that I sent email to 30 people yesterday and 900 people today. Good problem to have :) Please check your spam folder if you don't see today's issue in inbox.
* Make it easier to browse back in time. My biggest annoyance with news sites I've subscribed to is that they have no way to find articles after they have moved off the front page so I feel compelled to visit every day, and when I don't I feel like I'm wasting money, so I unsubscribe. I did eventually find that you have previous days available in the newsletter section, but a more discoverable interface like "previous/next day" link on the bottom of the page would be great.
* Add some less frequent newsletters, such as weekly and monthly which dedupe any stories that come up multiple times in that period, and include the top N stories, rather than everything over a threshold.
* Sections would be great for advanced users. My ideal would be to let the user set a different threshold for each section, but then still display them all together on the front page.
I am always interested in minimalist / text-oriented resources. And "importance ranking" seems like an especially valid task for AI. Thanks for sharing.
Any chance of updating the RSS feed item "summary" to something more useful than "significant news" and title to something more than the date? Or maybe an RSS feed of the newsletter format?
(I do see some of the older items have titles that include some info on what's included)
Basically, I think the ideal RSS feeds would be _two_:
1. RSS feed of every item that hits the 6/10 threshold (title and link to original, summary not too important, either same as title or paragraph summary of what the article is about)
2. RSS feed of the newsletter (title: "May 4 2023 - significant news", link: newsletter page, summary: list of article titles)
I suppose versions of #1 with diff thresholds could be nice, but I'd probably only use the default threshold.
Background: I run a bot that sends RSS feeds into discord channels (easily followable to any discord server) and this particular feed seems potentially quite handy for good news info. Probably overlaps with https://whatthefuckjusthappenedtoday.com/ but that automation probably means it can catch breaking news faster than a manually-curated site like that.
I've subscribed to the RSS feed. I thought the feed would provide links the significant news articles directly, but it appears to provide a link to the newsminimalist.com summary. That's totally fine, but I'd think it'd be a bit nicer if there was an option to access the articles directly.
I'm also using ChatGPT to scrape online info and reduce it. Were there any particularly interesting problems that you had to solve? For me, it was figuring out ways to reduce the number of tokens so I don't hemorrhage money!
Every news site has some notion of top items, but you’ve done it 10x better. The de-duplication of similar stories, great summaries, and super clean interface are amazing. Bravo! It really feels like you solved news aggregation.
I noticed some bias because I only analyze sources in English. It should be easy to make ChatGPT also read news in other languages, maybe that will help. Is that what you meant?
It also seems to be very US-centric and seems to put a strong focus on reporting on the economy. I personally found nothing particularly relevant to my existence featured on the site.
These two biases are probably indicative of a host of other, less obvious biases. To be clear: all media is biased, because it is created by biased humans and it has been well demonstrated that algorithms replicate the biases of their creators.
A good newspaper/website knows their audience and delivers reporting relevant to it in a language they appreciate and cognizant of its societal impact.
> It should be easy to make ChatGPT also read news in other languages
This should indeed be easy, and given how good it is at translation, very interesting - taking too news sources and national broadcast services; rate as you do now - filter the top, summarize in English, store and dedup on similarity vector - it would be great. For example i don't read any Chinese or African news sources - but it would be great to have them in the mix.
Might even index on similarity first then ask for a summary on all different reports on the same story?
A sentiment analysis plot of all the sources could be insightful both for users as well as a way to tune the weight of sources in the future. Very cool
This is awesome! I've used thefactual and ground news briefly in the past to try to stay informed without becoming a news-junkie-zombie or exposing myself to propaganda unnecessarily, but ultimately gave up on even "staying informed" because the political nonsense always leaked in and required manual filtering on my part. These headlines are the tiny little tidbits of information that I actually care about, and nothing else. Love that there's no commentary, no extra context, nothing. Just some headlines.
My take is that this has legs and please iterate on it to it's natural conclusion, whatever that looks like. It's executed well as is but seems sparse on features overall. Cool idea
I love this idea. For a while this has been the filtering function that HN provided for me (in addition to higher quality discourse) but this does it for general news.
6.5 seemed like a sweet spot for number of news. It results in 1-5 news each day. I tried 6.0 at some point, but I think it resulted in like 10-20 articles each day which seemed a bit much.
Yeah, sorry, should've made it earlier! Here's the default one from the newsletter platform, but some people had complaints about it: https://rss.beehiiv.com/feeds/4aF2pGVAEN.xml
Seems us/eu-centric? I’ll definitely keep an eye on it to see if the composition changes with important events elsewhere in the world.
Even if only English language sources are considered, there are available ones for non-English-speaking countries, example from Spain https://english.elpais.com/
News isn't significant, only relevant, as the significance is relative. Who's reading it, where they are, what their interests are, what impacts them, and so on. There's no way to generalize it because even some people who it directly impacts may simply not care; it's not significant for them. So the best way to deliver significant news is to personalize, not generalize.
I did it at first, but quickly realized that it's too expensive.
GPT-4 is 15 times more expensive for prompt and 30 times more expensive for completion than GPT-3.5-turbo. But I think GPT-3.5-turbo gives 90% as good summaries as GPT-4.
Very nice I have subscribed to test it out for a couple of weeks.
I would love a simple mobile app, it's just easy to tap your news app on a phone, ideally with a way to customize the notification timing. I have certain moments each day when I want to consume this type of content (breakfast, lunch break)
A small community driven comment section with Karma system would also be nice. I like to discuss news and it enables knowledgeable people to add some additional insights. See the dutch tech news site for inspiration. they explain their karma system here: https://tweakers.net/info/faq/karma/#tab:1-2
Why not open it in the mobile browser? Or put a reference for the website on your home screen? You could use newsletterify.com (disclaimer: author here) to read them in the browser, not your inbox.
Yes, I understood the idea of the formula. Rather, my question is - how does chatGPT know how to value those variables? It doesn't have a real understanding AFAIK.
It doesn't. But by the magic of insanely large training corpus, it does anyway. Similar to how it can "roll a die" or "pick a card"[1], the weighted model can do some (to me) mind bending stuff, just by answering "which tokens are likely to come next, in this given context" (which, in some ways, is the same as me trying to guess what you are going to say next; "Thank you for explaining"? (Actually chatgpt is much better at that game than me...).
[1] I tried it with gpt4:
> Pick a card.
>> As an AI language model, I am unable to physically pick a card for you. However, if you're referring to a card from a standard 52-card deck, I can randomly select one for you.
>> Your randomly selected card is the 7 of Hearts.
> Again.
>> Alright, let's randomly select another card for you.
>> Your new randomly selected card is the Jack of Diamonds.
I must remember to check this again on a day with some very significant news, because at the moment everything it's showing me is kinda below the bar of what I want to read.
It's a nice idea and looks like it's implemented well. My problem is probably that different things are significant to different people. Following that thread leads to filter bubbles, of course.
Could News Minimalist have an option to only show good news with some significance? I think that might be popular and quite useful for the mental health of lots of people.
Good things do happen but they're under-reported because death, destruction and conflict draw more clicks & views. A way to surface the less depressing news would be welcome.
Seems like this is just another algorithmic feed? They're great at trimming down large amounts of data, but I always feel like algorithmic feeds always miss something, that a human never would, as I do just now reading the front page.
This seems to be highly focused on global financial news, which are unsignificant to most people on this planet.
It would be nice to be able to include or exclude certain news-categories to actually produce a list that has some significance to the reader.
I'm guessing the Arithmetic to maximising significance to population favours economics because 10 cents for every Indian outweighs 300 dead in a train crash several times over.
Indeed. The perfect news feed has to be tailored to the reader. For instance, events in my city have rather large impact on me but are not interesting for most of the planet.
I built a news app in the past that just read the headline news from a news source of every country on the planet (well, ones we could find English news sources for anyway), and then presented this in a single page where news from every country was equally represented.
You’d get super interesting contrasts, and when western media would be going wild over the latest Trump gaffe, right next to it there’d be news from some island in the middle of pacific saying their antelope conservation efforts were finally paying off.
In hindsight it was super cool, and I’m really sad I didn’t keep up with it.
231 comments
[ 3.1 ms ] story [ 343 ms ] thread>“Today ChatGPT read 1289 top news and gave 13 of them a significance score over 6/10.”
Is an excellent hook.
I wish there was more of a basis on the score it chose. For example:
>“Russia suffers 100,000 casualties in Ukraine conflict, US estimates.” is #2 and ranked 6.8
>“White House estimates Russia has suffered 100,000 casualties in Ukraine since December.” is #237 and ranked 3.8.
Initially I asked ChatGPT to estimate three things: event scale, event magnitude and event potential. That often resulted in clickbait articles going to the top.
To fix this I started to also ask it to estimate source credibility, so tabloids would get much lower score than, say New York Times.
Now you noticed another problem, similar articles get very different scores. I think ideally I could do some sort of deduplication, but I don't know how to implement it yet.
Any chance AI could be used to dedup the stories (like these are identical - only show higher source)?
The problem with deduping is that some news get posted and reposted by different sources for several days (sometimes even weeks) in a row. That's a huge context I'd have to put in AI.
1000 news titles * 3 days * 70 symbols per title = 210,000 symbols = 40000 words = 53000 tokens. My current context window is 8000 tokens and I think 32000 tokens is max that GPT-4 allows.
---
Add: now that I think about it should be possible to do in several runs. Will keep thinking about it, thanks for the suggestion.
Every time ChatGPT saw the words "World is ending" (not real example) it gave those articles very high score.
Estimating source credibility was the only solution I came up with.
100,000 dead Russians, is very likely the propaganda number. Maybes it’s right; maybe it’s not, IDK, but I know that there has been zero negative Ukraine news since the conflict started (ghost of Kiev anyone?), and the media machine definitely only works one way on this topic.
I mean, Russia isn’t the good guy here, but I don’t excuse propaganda just because it tells a story I want to hear.
In this case, I don’t care what the US or Ukraine says the number is. That isn’t news. It’s narrative, true or not.
So, I feel like the score should take into account if how likely an article is to be narrative vs purely an event.
The number is news because Russia is waging a large-scale invasion of another country using cannon fodder tactics. Reporting on that invasion is not simply propaganda.
Since December
Happy to answer any questions.
(below is the project description I used when posted about it on Reddit)
The problem I have with most news sites is that I can't read only important news: an article about a virus outbreak is followed by some celebrity gossip or another smartphone release.
But even on sites that focus on important events articles are posted every day and there are always "top headlines" — even on days when nothing important happened.
I am forced to make a choice: waste time going through unimportant updates or ignore the news and miss important events.
So I built a web app that I think solves this.
It uses AI (ChatGPT-4) to read the top 1000 news every day and rank them by significance on a scale from 0 to 10 based on event magnitude, scale, potential, and source credibility.
The results are posted on the site: https://www.newsminimalist.com/
I also run a newsletter where I post summaries of all the news with a score over 6.5. On average that's 1-3 articles per day, but sometimes it is 5, and sometimes — none at all. In that case, I just send an email saying that nothing important happened that day.
You can read previous issues here: https://newsletter.newsminimalist.com/
Let me know if you have any feedback or ideas. I'm considering adding new features and looking for direction.
Eg default (no topic sections like what you have today), and optional advanced (with sections)
( I used ChatGPT to help me find it, via https://smartynames.com/ )
Thank you kind stranger!
Question: are you using ChatGPT to summarize the article and making that the title of each post?
Relatedly, it would be neat if you had a slider or +/- buttons for each article summary, so someone could choose their depth based on how interesting the title sounds.
I just ask chatgpt to "summarize the article very concisely" and hope that it won't lose any important points. But chatgpt is really good at summarizing, so I don't worry much.
Having +/- buttons is a good idea, will add it to todo list as well.
If you were so inclined, it would probably be pretty trivial to also offer leveled versions as well, which would be very useful for schools/teachers to use with students. There are a handful of companies that have bespoke leveled reading articles created from standard news articles. ChatGPT would make the workflow for creating news articles much easier, though manual QC would still be necessary to ensure that the article summaries are accurate and appropriate for students.
Recent example: https://newsletter.newsminimalist.com/p/tuesday-april-25-3-m...
Heads up that your plaintext emails in the newsletter have some html formatting and template artifacts in there.
Adding filter for news sources should be pretty easy, I'll consider it for "advanced view".
Now that would be worth something...
Say, you want more than 6 filters they costs 50 cents/filter/month with a minimum of 5.
I think one would gradually make more and more of them?
Have you made any attempt at quantifying its biases? Both whether it considers left- or right-leaning articles more significant and where it focuses most of its reporting. (Economy and the US from a first glance?)
> Have you made any attempt at quantifying its biases?
I haven't. I can share ChatGPT's evaluation of sources credibility, maybe that will give people some insights. For example, Reuters got 9.5 (out of 10), Lifehacker 7, Oprah Mag 3.
> where it focuses most of its reporting. (Economy and the US from a first glance?)
I'd say today is just US and economy-heavy day. There were many days when other categories and geographies made the top (I especially enjoy these). One example: https://newsletter.newsminimalist.com/p/tuesday-april-25-3-m...
FWIW, when I read news I often cross check on different outlets to get a sense of the span, range and sometimes media blackouts, including the ones with extreme bias (like RT reporting on Ukraine). This can give a good reading of the meta-temperature of a developing or controversial situation with a lot of propaganda and bias. In some cases, less trustworthy outlets will cover stories that aren’t narrative-friendly to more reputable publications, and sometimes, those can be really important.
For a service like this, I’d much prefer something analogous to “here’s what different outlets are saying”, rather than trying to make its own judgment about bias.
Subscribed to newsletter.newsminimalist.
Personally, I want an even higher signal to noise ratio and even fewer articles. Perhaps significance > 7, and articles from the last week.
I actually created a site - https://detoxed.news - with a very similar philosophy a while back, though with a much simpler implementation. It periodically scrapes Wikipedia's Current Events Portal (https://en.wikipedia.org/wiki/Portal:Current_events) and presents that information in a nicer format.
I think there's definitely a sweet spot where you keep well-informed about current events and happenings in the world, without wasting your time on the 24 hours new cycle.
1) Flood in Rwanda kills 109 people 2) Sudan War 3) Australia evacuation from Sudan 4) Seizing of an oil tanker
None of these headlines keep me particularly informed and are actually a more potent version of the 24 hours new cycle
[0] https://cln.sh/s44CS9tmwkC79NHcXPP3
The idea of filtering news sourced mentioned elsewhere is nice, but will necessitate considerable input from users, kind of negating the purpose of the site
Have you considered/thought about this aspect of news?
I bet spam algorithms didn't like that I sent email to 30 people yesterday and 900 people today. Good problem to have :) Please check your spam folder if you don't see today's issue in inbox.
* Make it easier to browse back in time. My biggest annoyance with news sites I've subscribed to is that they have no way to find articles after they have moved off the front page so I feel compelled to visit every day, and when I don't I feel like I'm wasting money, so I unsubscribe. I did eventually find that you have previous days available in the newsletter section, but a more discoverable interface like "previous/next day" link on the bottom of the page would be great.
* Add some less frequent newsletters, such as weekly and monthly which dedupe any stories that come up multiple times in that period, and include the top N stories, rather than everything over a threshold.
* Sections would be great for advanced users. My ideal would be to let the user set a different threshold for each section, but then still display them all together on the front page.
Any chance of updating the RSS feed item "summary" to something more useful than "significant news" and title to something more than the date? Or maybe an RSS feed of the newsletter format?
(I do see some of the older items have titles that include some info on what's included)
Basically, I think the ideal RSS feeds would be _two_: 1. RSS feed of every item that hits the 6/10 threshold (title and link to original, summary not too important, either same as title or paragraph summary of what the article is about) 2. RSS feed of the newsletter (title: "May 4 2023 - significant news", link: newsletter page, summary: list of article titles)
I suppose versions of #1 with diff thresholds could be nice, but I'd probably only use the default threshold.
Background: I run a bot that sends RSS feeds into discord channels (easily followable to any discord server) and this particular feed seems potentially quite handy for good news info. Probably overlaps with https://whatthefuckjusthappenedtoday.com/ but that automation probably means it can catch breaking news faster than a manually-curated site like that.
"RSS" that I have currently is generated by the email platform that I use for daily emails. It's not a proper RSS from website.
Making a proper RSS for website items is on my list, but I need to finish some other features first.
Agree about subtitles in older items, I guess I just got lazy writing them :) your message was just what I needed to get back to older better format.
I've subscribed to the RSS feed. I thought the feed would provide links the significant news articles directly, but it appears to provide a link to the newsminimalist.com summary. That's totally fine, but I'd think it'd be a bit nicer if there was an option to access the articles directly.
I'm also using ChatGPT to scrape online info and reduce it. Were there any particularly interesting problems that you had to solve? For me, it was figuring out ways to reduce the number of tokens so I don't hemorrhage money!
Yeah RSS it not great currently. Replied about it here with more info: https://news.ycombinator.com/item?id=35820032
Oh, exactly the same problem. The straightforward implementation would make me broke in month :)
Had to make several optimizations to make the bill reasonable. Not all of them worked, some greatly reduced the quality of significance estimations.
Reducing number of tokens was super fun, forgot about it :D
These two biases are probably indicative of a host of other, less obvious biases. To be clear: all media is biased, because it is created by biased humans and it has been well demonstrated that algorithms replicate the biases of their creators.
A good newspaper/website knows their audience and delivers reporting relevant to it in a language they appreciate and cognizant of its societal impact.
This should indeed be easy, and given how good it is at translation, very interesting - taking too news sources and national broadcast services; rate as you do now - filter the top, summarize in English, store and dedup on similarity vector - it would be great. For example i don't read any Chinese or African news sources - but it would be great to have them in the mix.
Might even index on similarity first then ask for a summary on all different reports on the same story?
I have no control over it, but let me know if there are any problems with it — I'll see what I can do.
That is, for a day with 5 news, post 5 times on the feed, one for each.
I'm using a RSS reader that only shows the title of each day (and not the body of the post), and right now it's showing only
> Tuesday, May 2 — 4 significant news
> Only significant news. All signal, no noise.
But I would like to be able to skim all headlines in the RSS reader itself
(I know I could probably make a custom RSS feed out of this though)
Kudos!
0.5/10 might be the sweet spot for best of the worst.
- WWE Draft 2021: Triple H announces draft picks and new brand exclusivity.
- Kim Kardashian and Pete Davidson reunite at 2023 Met Gala.
- Vladimir Putin's lover Alina Kabaeva makes rare appearance at gymnastics event in Siberia.
I’d love to see some dials to find even worse news in the future. :)
Q: Could you please add RSS? I’d use this every day if I could get it in Feedly. (NM found it in a comment https://rss.beehiiv.com/feeds/4aF2pGVAEN.xml)
Q: Why’d you choose 6.5 as the default? Sorry if you answered that already.
I will definitely sub to News Minimalist if the same is possible.
Will try to improve it.
Added "good RSS" to the TODOs.
Even if only English language sources are considered, there are available ones for non-English-speaking countries, example from Spain https://english.elpais.com/
It might not be for much longer though...
As you said analyzing news in other languages should help with bringing up different perspectives. I'll work on that.
But I'm still not sure what to do about chatgpt's bias. For now I just accept it.
I did it at first, but quickly realized that it's too expensive. GPT-4 is 15 times more expensive for prompt and 30 times more expensive for completion than GPT-3.5-turbo. But I think GPT-3.5-turbo gives 90% as good summaries as GPT-4.
I would love a simple mobile app, it's just easy to tap your news app on a phone, ideally with a way to customize the notification timing. I have certain moments each day when I want to consume this type of content (breakfast, lunch break)
A small community driven comment section with Karma system would also be nice. I like to discuss news and it enables knowledgeable people to add some additional insights. See the dutch tech news site for inspiration. they explain their karma system here: https://tweakers.net/info/faq/karma/#tab:1-2
> For each article ChatGPT estimates:
> Magnitude: how big was the effect;
> Scale: how many people the event affected;
> Potential: how likely is it that the event will cause bigger events;
> Credibility: how credible is the source.
Then I do `Credibility * cbrt(Magnitude * Scale * Potential)`
cbrt = cubic root
Maybe the formula could be improved but I was satisfied with the result, so kept it that way.
[1] I tried it with gpt4:
> Pick a card.
>> As an AI language model, I am unable to physically pick a card for you. However, if you're referring to a card from a standard 52-card deck, I can randomly select one for you.
>> Your randomly selected card is the 7 of Hearts.
> Again.
>> Alright, let's randomly select another card for you.
>> Your new randomly selected card is the Jack of Diamonds.
It's a nice idea and looks like it's implemented well. My problem is probably that different things are significant to different people. Following that thread leads to filter bubbles, of course.
For big international (usually bad!) news I tend to go to Wikipedia Current Events: https://en.wikipedia.org/wiki/Portal:Current_events
Could News Minimalist have an option to only show good news with some significance? I think that might be popular and quite useful for the mental health of lots of people.
Good things do happen but they're under-reported because death, destruction and conflict draw more clicks & views. A way to surface the less depressing news would be welcome.
Or perhaps they are more significant than they feel because their trickle-down effect is slow and hard to trace.
You’d get super interesting contrasts, and when western media would be going wild over the latest Trump gaffe, right next to it there’d be news from some island in the middle of pacific saying their antelope conservation efforts were finally paying off.
In hindsight it was super cool, and I’m really sad I didn’t keep up with it.