https://en.wikipedia.org/wiki/Portal:Current_events
Although that's perhaps way too minimalist?
I was disagreeing a lot with their selection of news, for example one of their recent entries is:
"Two people are killed and eleven others are injured when a bus flips on its side on a highway near Prenzlau, northeast of Berlin, Germany."
My system gave it a significance score of 1.8, so similar news should never get to the main page: https://www.newsminimalist.com/articles/two-dead-and-four-in...
A lot of the most signficant stories are political, for example, which someone may have no interest in.
I have had this same idea in the past, tuning to my personal interests.
I separate significance from importance (or relevance).
Importance is subjective. News about the health of my family members is important to me, but it is not significant to the world.
Significance is objective. It's about how much the event affects humanity as a whole.
I don’t agree with that, at least not in the present. We only know what’s truly significant when we reflect on history. There are very few things we can be certain are significant in the present. Climate change is likely one, but the US debt ceiling and the war in Ukraine don’t seem as likely to me, at least not in the human scale. There are also events that happen that don’t appear significant in the present but will be hugely significant in the future.
My goal with this project is not to get "true" significance but to have a setup that gets you 90% of the way there: an automated system that finds events that are likely to affect large groups of people or major systems and filters out most of everyday noise.
There will always be false-positives and false-negatives, but I think it's a good starting point and it should slowly get better as models get smarter.
$ wget -qO - https://rss.beehiiv.com/feeds/4aF2pGVAEN.xml | grep pubDate
<pubDate>Thu, 16 Jan 2025 17:56:47 +0000</pubDate>
<pubDate>Thu, 16 Jan 2025 17:56:47 +0000</pubDate>
<pubDate>Mon, 13 Jan 2025 17:22:34 +0000</pubDate>
<pubDate>Sat, 11 Jan 2025 16:42:01 +0000</pubDate>
<pubDate>Thu, 09 Jan 2025 17:52:38 +0000</pubDate>
<pubDate>Fri, 03 Jan 2025 16:19:25 +0000</pubDate>
<pubDate>Sat, 28 Dec 2024 17:44:35 +0000</pubDate>
<pubDate>Wed, 18 Dec 2024 16:44:00 +0000</pubDate>
<pubDate>Mon, 16 Dec 2024 18:01:56 +0000</pubDate>
<pubDate>Tue, 10 Dec 2024 18:00:55 +0000</pubDate>
<pubDate>Fri, 06 Dec 2024 17:43:37 +0000</pubDate>
<pubDate>Wed, 04 Dec 2024 16:38:38 +0000</pubDate>
<pubDate>Sun, 01 Dec 2024 18:57:43 +0000</pubDate>
<pubDate>Fri, 29 Nov 2024 18:07:14 +0000</pubDate>
<pubDate>Sun, 24 Nov 2024 15:26:59 +0000</pubDate>
<pubDate>Thu, 21 Nov 2024 17:25:22 +0000</pubDate>
<pubDate>Tue, 19 Nov 2024 17:47:19 +0000</pubDate>
<pubDate>Fri, 15 Nov 2024 17:28:00 +0000</pubDate>
<pubDate>Tue, 12 Nov 2024 17:44:08 +0000</pubDate>
<pubDate>Fri, 08 Nov 2024 17:22:01 +0000</pubDate>
<pubDate>Wed, 06 Nov 2024 16:43:10 +0000</pubDate>
Also: It'd be great if you had a feed tag in your HTML head, so RSS readers could pick it up straight out of your homepage URL instead of needing to manually hunt for the right RSS link.
source: I run inference locally and built the server for around $6k. I get upwards of 10t/s on deepseek v3
PS: thank you for running this service. I've been using it casually since launch and find it much better for my mental health than any other source of news I've tried in the past.
There was a period when I considered switching to an open-source model, but every time I was ready for a switch, OpenAI released a smarter and often cheaper model that was just too good to pass up.
Eventually I decided that the potential savings are not worth it in the long term - it looks like LLMs will only get cheaper over time and the cost of inference should become negligible.
Or, even farther off the deep-end: have you considered open-sourcing any old versions of your prompts or pipeline? Say one year after they are superseded in your production system?
I had a very similar idea a while back. I wanted to rank news by "impact" which might be more concrete than "significance."
For an LLM prompt, it would be something like:
"estimate the number of people who's lives that will be materially changed by this news." and "estimate the average degree of change for those impacted."
Then impact is roughly the product of those two.
Additionally, I want a version that is tailored to me specifically "estimate the degree of change this will have on my life." + context of my life.
Tangentially, I've found that getting ratings out LLMs works better when I can give all options and request relative ratings. If I ask for rankings individually I get different and less good results. Not enough context length to rate all news from all time in one go though. Any thoughts on that? Maybe providing some benchmark ratings with each request could help? Something I'm exploring.
In the beginning I had 3 parameters: scale (number of people), magnitude (degree of change for those impacted) and additionally potential (how likely is this event to trigger downstream significant events).
The point behind including potential was to separate these two events:
1) A 80 year old dies from cancer 2) An 80 year old dies from a new virus called COVID
This worked roughly well but I kept adding parameters to improve the system: novelty, credibility, etc... The current system works on 7 parameters.
---
I never attempted to give LLM all options and rank them against each other.
1) as you said, for me 20k articles is just too much to fit into context window. Maybe some modern LLMs can handle it, but it wasn't the case for a long time, and I settled on current approach.
2) I don't want the "neighbors" to affect individual article ratings. With the current system I am able to compare news spread over months, because they were all rated using the same prompt.
3) I intentionally avoided giving AI examples, like "evaluate event X given that event Y is 7/10". I want it to give scores with a "clear mind" and not be "primed" to my arbitrary examples.
I'm working on something similar but am thinking of using AI differently. Great job Vadim.
It's available via RSS too: https://rss.beehiiv.com/feeds/4aF2pGVAEN.xml
I attempted to make a weekly version, but quickly dropped the idea. Over the course of the week articles often became outdated (not just old, but plain wrong).
I found that an optimal newsletter schedule is sending it about every 48-72 hours, depending on how eventful that period was. With this frequency, the articles rarely become outdated, and at the same time it's not too frequent to get tired of.
Might take a stab at ranking them as well.
"ChatGPT, set a daily/weekly task to give me the most significant news. Use this ranking criteria: <input criteria>"
I think LLMs are really underutilized as a "judgement tool". A couple similar ideas people reached out to me with were: evaluating which pull requests are more significant in a big repo, or which grant applications have more merit.
The LLMs will always make mistakes, but they could work great as the first filter.
As in every other engineering endeavor, the raw data you start off with has a lot to do with what you end up with, no matter what transforms happen. :)
Wikipedia current events page was actually one of the reasons for creating this project! I was disagreeing a lot with their selection of news, for example one of the recent entries is:
"Two people are killed and eleven others are injured when a bus flips on its side on a highway near Prenzlau, northeast of Berlin, Germany."
My system gave it a significance score of 1.8: https://www.newsminimalist.com/articles/two-dead-and-four-in...
Most other aggregators show news based on 1) relevance, 2) upvotes or 3) coverage.
Relevance-based algorithms tend to put reader into a bubble, where the more they read on a certain topic, the more news they see on that topic.
Upvotes-based algorithms usually bring up a lot of clickbait and drama.
Sorting by coverage doesn't really work either, media often just follows people's interests and churns articles on what is "hot".
For example, last summer, a fight between Zuckerberg and Musk was at the top of most feeds based both on upvotes and coverage. Significance-based algorithm didn't even put it in the top 50.
No plans for an iOS app, but my site is a PWA so can be installed and look like an app both on Android and iOS.
How can I filter out (block) subjects/words?
"World" includes everything that talks about two or more countries.
"Nation" is a category with news that touch only a single country, not necessarily US. The current feed is very US-heavy because practically everyone (even non-US sources) is talking about Trump.
Keyword blocking is available on premium: https://www.newsminimalist.com/premium#block-topics
The default feed sorting is done for regular visitors (new first), for evaluating the output you might like the "significant first" more: https://www.newsminimalist.com/?sort=significance
On that list, the ceasefire article is on the second place out of the ~40k articles analyzed.
---
Having more variables is an good idea. I don't have an immediate vision on how to use it in the UI (I want to keep it minimal), but will think more about it.
---
I've been really torn on free trial. I currently offer a refund guarantee, but will add a trial as well soon.
But from what I've seen it's pretty indifferent to "sides", it's more focused on raw numbers of people affected and magnitude of the event.
rank by significance:
NOT politics; NOT Trump; NOT republican; NOT democrat