Interesting or Not is a ‘neo-news platform’ that ranks news and information in terms of concepts, ideas, and models that you will find interesting—like long-termism.
In short, it lets you easily read news that you find interesting – rather than what a publisher tells you is, or should, be interesting.
Most news isn’t that interesting or high quality. That is, at least in part, a by-product of how traditional mainstream media allocates information about the world, which is generally through proxies of what they think people will find interesting—national politics, celebrity gossip, and catch-all social commentary, for instance.
Information (news) allocation, and who gets what they want to read, is treated like an unscalable artisanal craft that only people with honed expertise are ostensibly able to do. Yet, in reality, it doesn’t have to be like that. Instead, it’s a technical problem that can be solved with technology. When you stop viewing the problem as “what do we think people want to read today,” but instead as “what’s the most effective way to give people what they want to read,” you get very different answers.
It’s reasonable to assume, therefore, that unless they are telling you otherwise, mainstream media outlets are ranking the order in which they publish news arbitrarily. It may seem intuitively consistent, but that of course doesn’t mean that the way they rank news is good. It’s safe to assume that news is currently ranked in terms of what memes currently have widespread capture—many memes are bad, dangerous, and just uninteresting, so that’s hugely concerning!
We think that news should be ranked consistently and with the reader directly in mind. People should be able to read about what they find interesting!
Short-termist news
Importantly, most news isn’t written thoughtfully with the long-term in mind. Instead, the obvious cynical interpretation of what motivates news writing is that it is motivated by what will sell newspapers or clicks. This perpetuates short-termist news cycles that are generally pernicious to any serious discussion of a topic.
In the UK at the moment there is an election taking place. It’s hard to characterise what is taking place at the moment as anything other than very short-termist, with a focus on who will win in a few weeks time, rather than serious discussion about important issues. I’ve seen no discussion of serious long-term issues like AI safety (and how to harness its potential), how to minimise X-risks/S-risks (which is astonishing given recentness of Covid), and on a meta-level how to improve the operation of government.
All of this is ultimately very pernicious, and means that no serious discussion of important long-term issues: how do we prevent a future pandemic? How do we reduce the risk of nuclear war/war in general? How do we ensure AI safety whilst harnessing its potential?
Interest silos
People are quite reasonably interested in different things—and so they should be! Epistemic diversity of this sort is a net good and means that people can approach similar problems from different angles.
Mainstream news at the moment approaches news in a relatively homogenous fashion. It generally talks about similar events in a similar way, with the only variable generally being whether its left/right politically. There are exceptions of course. The opportunity to talk about events and ideas in completely diverse ways is, therefore, huge!
Now there are of course legitimate concerns about falling into entrenched political silos. That is, if anything, what Interesting or not is trying to avoid. People shouldn’t fall into a binary of silos, or indeed a limited set of silos. Rather we should all have the choice of a multiplicity of idea silos. Richer mental frames and models is epistemically beneficial.
How does it work?
It uses (the same tech as) spam filters to give news articles a score of how much it is like an idea/concept/mental model.
Interesting or Not uses machine learning text classifiers to rank how much one piece of text is like another. The long-termist model, for example, is trained on articles from ~30 years ago that are either i) lastingly interesting today, or ii) not interesting today.
With 100 articles of each category, we generate an approximate long-termist model of lastingly interesting news, which we then use to rank today’s news.
Where did the idea come from?
Paul Graham, the co-founder of YCombinator, is, among other things, well known for coming up with a simple, yet brilliant, solution to removing spam emails from email inboxes. He’s written several essays on his work in this area.
More recently, however, he’s started to toy with the idea of applying the tech behind spam filters to other domains, like news classification. In 2021, Graham tweeted:
Mark news stories from some past decade (e.g., 90s) as still interesting or not.
Train a text classifier on the resulting two corpora.
Sort today’s stories according to their score.
If there was an aggregator that did this, I’d probably visit it every day.
And then again in 2024, Graham tweeted:
Idea I wish someone would try: train a text classifier on NYT stories from a couple decades ago by dividing them into two corpora, lastingly important and not, and then see if you can use this to predict which stories published today are lastingly important.
Working close-by to where a lot of news is generated, I’ve seen how unrealistic, or often wrong, a lot of news reporting is. So, I was very keen to try and see if you could alter the quality of news by intervening in the allocation process.
But, also news (perhaps because of our availability bias) is harshly skewed towards short-termism, and not long-termism. This seemed to me to be hugely counter-productive! Think about some of the last news headlines that you’ve seen and they are probably about short-term political events, rather than long-term serious issues.
Orientating news towards long-term discussion, and actually important issues, and away from the latest gaffe/crisis/gossip/scandal would be of serious epistemic value!
Goals
Personalized news feeds
Currently, there are the lightbulb buttons and the disabled ‘For You’ label at the top of the feed. Our next goal is to make these active, so that you can generate your own personal model of what news is ‘For You’/what you find interesting.
You should have your own model of what news you find interesting.
More models and more metrics
The platform currently has four metrics that it ranks news by:
The Long-Term Score (how long-termist is the article)
The Hacker News Score (how like Hacker News is the article)
The Marxist Score (how Marxist is the article)
The SW1 Politics Score (news that will appeal to UK political types)
These are just a few that we’ve come up with so far, but there is an endless number of models/filters that we could add. For instance, an Effective Altruist filter or a rationalist (Lesswrong) filter. We will add more in the coming weeks.
But it’s not just these complex and idea-based metrics that we think will prove interesting. News text includes a whole host of interesting mineable information. For example, sentiment, political leaning, and geo-distribution. These are all metrics that can be mined from text. This should mean that we can eventually give you the metrics of what your ‘information diet’ looks like.
Do you read more left or right-wing content? Is the news you read more positive or negative (how susceptible are you to your negativity bias)? What concepts, ideas, and topics do you prefer? This is all quantifiable information that we’re looking to give to users.
A platform for both readers and writers
We want Interesting or Not to not just be a news aggregator. Instead, we want it to be a platform where both readers can come to read what they find interesting and useful, but also where writers can come to reach their audiences—that they wouldn’t otherwise be able to find.
That is why we’re going to be building a back-end for writers to publish their content so that it can reach readers that want it.
Feedback
We’re still very much in the MVP stage of Interesting or Not, so any feedback, good, bad, critical, or not, would be hugely appreciated!
What is one thing that you would change about the site?
Or some other areas that we would really appreciate comments on:
Is this a problem that you have—not being able to easily find news that you like/find interesting?
Would you use a platform that accurately gives you personalized news you find interesting?
Do you find the current concept lenses interesting and useful?
What (RSS) outlets could we add for you, and what filters would you like added?
Do you like the User Interface, or could we present the news in another way?
Would you like metrics on the news that you consume?
Or, even if you just want to let us know that this is a problem you have, that’d be appreciated too!
I’m building a long-termist ‘neo-news’ platform
Interesting or Not
Interesting or Not is a ‘neo-news platform’ that ranks news and information in terms of concepts, ideas, and models that you will find interesting—like long-termism.
In short, it lets you easily read news that you find interesting – rather than what a publisher tells you is, or should, be interesting.
Please see our current MVP here: interestingornot.com
And, find us on Twitter/X: x.com/SIONFUTURE
What problem does it solve?
Low quality uninteresting news
Most news isn’t that interesting or high quality. That is, at least in part, a by-product of how traditional mainstream media allocates information about the world, which is generally through proxies of what they think people will find interesting—national politics, celebrity gossip, and catch-all social commentary, for instance.
Information (news) allocation, and who gets what they want to read, is treated like an unscalable artisanal craft that only people with honed expertise are ostensibly able to do. Yet, in reality, it doesn’t have to be like that. Instead, it’s a technical problem that can be solved with technology. When you stop viewing the problem as “what do we think people want to read today,” but instead as “what’s the most effective way to give people what they want to read,” you get very different answers.
It’s reasonable to assume, therefore, that unless they are telling you otherwise, mainstream media outlets are ranking the order in which they publish news arbitrarily. It may seem intuitively consistent, but that of course doesn’t mean that the way they rank news is good. It’s safe to assume that news is currently ranked in terms of what memes currently have widespread capture—many memes are bad, dangerous, and just uninteresting, so that’s hugely concerning!
We think that news should be ranked consistently and with the reader directly in mind. People should be able to read about what they find interesting!
Short-termist news
Importantly, most news isn’t written thoughtfully with the long-term in mind. Instead, the obvious cynical interpretation of what motivates news writing is that it is motivated by what will sell newspapers or clicks. This perpetuates short-termist news cycles that are generally pernicious to any serious discussion of a topic.
In the UK at the moment there is an election taking place. It’s hard to characterise what is taking place at the moment as anything other than very short-termist, with a focus on who will win in a few weeks time, rather than serious discussion about important issues. I’ve seen no discussion of serious long-term issues like AI safety (and how to harness its potential), how to minimise X-risks/S-risks (which is astonishing given recentness of Covid), and on a meta-level how to improve the operation of government.
All of this is ultimately very pernicious, and means that no serious discussion of important long-term issues: how do we prevent a future pandemic? How do we reduce the risk of nuclear war/war in general? How do we ensure AI safety whilst harnessing its potential?
Interest silos
People are quite reasonably interested in different things—and so they should be! Epistemic diversity of this sort is a net good and means that people can approach similar problems from different angles.
Mainstream news at the moment approaches news in a relatively homogenous fashion. It generally talks about similar events in a similar way, with the only variable generally being whether its left/right politically. There are exceptions of course. The opportunity to talk about events and ideas in completely diverse ways is, therefore, huge!
Now there are of course legitimate concerns about falling into entrenched political silos. That is, if anything, what Interesting or not is trying to avoid. People shouldn’t fall into a binary of silos, or indeed a limited set of silos. Rather we should all have the choice of a multiplicity of idea silos. Richer mental frames and models is epistemically beneficial.
How does it work?
It uses (the same tech as) spam filters to give news articles a score of how much it is like an idea/concept/mental model.
Interesting or Not uses machine learning text classifiers to rank how much one piece of text is like another. The long-termist model, for example, is trained on articles from ~30 years ago that are either i) lastingly interesting today, or ii) not interesting today.
With 100 articles of each category, we generate an approximate long-termist model of lastingly interesting news, which we then use to rank today’s news.
Where did the idea come from?
Paul Graham, the co-founder of YCombinator, is, among other things, well known for coming up with a simple, yet brilliant, solution to removing spam emails from email inboxes. He’s written several essays on his work in this area.
More recently, however, he’s started to toy with the idea of applying the tech behind spam filters to other domains, like news classification. In 2021, Graham tweeted:
Mark news stories from some past decade (e.g., 90s) as still interesting or not.
Train a text classifier on the resulting two corpora.
Sort today’s stories according to their score.
If there was an aggregator that did this, I’d probably visit it every day.
And then again in 2024, Graham tweeted:
Idea I wish someone would try: train a text classifier on NYT stories from a couple decades ago by dividing them into two corpora, lastingly important and not, and then see if you can use this to predict which stories published today are lastingly important.
Working close-by to where a lot of news is generated, I’ve seen how unrealistic, or often wrong, a lot of news reporting is. So, I was very keen to try and see if you could alter the quality of news by intervening in the allocation process.
But, also news (perhaps because of our availability bias) is harshly skewed towards short-termism, and not long-termism. This seemed to me to be hugely counter-productive! Think about some of the last news headlines that you’ve seen and they are probably about short-term political events, rather than long-term serious issues.
Orientating news towards long-term discussion, and actually important issues, and away from the latest gaffe/crisis/gossip/scandal would be of serious epistemic value!
Goals
Personalized news feeds
Currently, there are the lightbulb buttons and the disabled ‘For You’ label at the top of the feed. Our next goal is to make these active, so that you can generate your own personal model of what news is ‘For You’/what you find interesting.
You should have your own model of what news you find interesting.
More models and more metrics
The platform currently has four metrics that it ranks news by:
The Long-Term Score (how long-termist is the article)
The Hacker News Score (how like Hacker News is the article)
The Marxist Score (how Marxist is the article)
The SW1 Politics Score (news that will appeal to UK political types)
These are just a few that we’ve come up with so far, but there is an endless number of models/filters that we could add. For instance, an Effective Altruist filter or a rationalist (Lesswrong) filter. We will add more in the coming weeks.
But it’s not just these complex and idea-based metrics that we think will prove interesting. News text includes a whole host of interesting mineable information. For example, sentiment, political leaning, and geo-distribution. These are all metrics that can be mined from text. This should mean that we can eventually give you the metrics of what your ‘information diet’ looks like.
Do you read more left or right-wing content? Is the news you read more positive or negative (how susceptible are you to your negativity bias)? What concepts, ideas, and topics do you prefer? This is all quantifiable information that we’re looking to give to users.
A platform for both readers and writers
We want Interesting or Not to not just be a news aggregator. Instead, we want it to be a platform where both readers can come to read what they find interesting and useful, but also where writers can come to reach their audiences—that they wouldn’t otherwise be able to find.
That is why we’re going to be building a back-end for writers to publish their content so that it can reach readers that want it.
Feedback
We’re still very much in the MVP stage of Interesting or Not, so any feedback, good, bad, critical, or not, would be hugely appreciated!
What is one thing that you would change about the site?
Or some other areas that we would really appreciate comments on:
Is this a problem that you have—not being able to easily find news that you like/find interesting?
Would you use a platform that accurately gives you personalized news you find interesting?
Do you find the current concept lenses interesting and useful?
What (RSS) outlets could we add for you, and what filters would you like added?
Do you like the User Interface, or could we present the news in another way?
Would you like metrics on the news that you consume?
Or, even if you just want to let us know that this is a problem you have, that’d be appreciated too!
Thanks for reading if you got this far!
[N.B. very much inspired by Vandemonian’s post on the Base Rate Times!]