Project update + 2 significant news stories

Hi there! As always, read the project update after the news.

Today ChatGPT read 12523 top news stories. After removing previously covered events, there are 2 articles with a significance score over 5.9.

[6.0] Trump threatens Brics nations with 100% tariffs if they undermine dollar — ft.com [$]

US president-elect Donald Trump has threatened to impose 100% tariffs on Brics nations if they pursue a new currency to rival the US dollar. He demands these countries commit to not creating or supporting an alternative currency.

The Brics group includes Brazil, Russia, India, China, South Africa, and others. Some members have expressed interest in challenging the dollar's dominance, which Trump aims to prevent.

[6.1] Severe flooding in Malaysia displaces over 120,000 — bbc.com

Heavy rain in Malaysia has caused severe flooding, displacing over 122,000 people and resulting. The situation is particularly critical in Kelantan, where 63% of evacuees are located. The National Disaster Management Agency has set up 679 shelters to assist those affected.

In neighboring Thailand, six provinces have also declared a disaster, impacting over 240,000 households. The army is assisting with rescue efforts as both countries face challenges during the monsoon season.

Highly covered news with significance over 5.5

[5.6] Putin approves record military budget for 2025
(apnews.com + 10)

[5.5] UN halts aid deliveries through Gaza's main crossing due to looting concerns
(apnews.com + 161)

Project update

It’s been almost half a year since my last update! I have a lot to share, but will try to be short today.

First, I think I finally got close to “solving” the scoring.

Since the beginning of the project, the scoring was done in one run: ChatGPT saw the news for the first time and gave its scores instantly, without ‘thinking’. It’s like asking someone to answer a question with the first thing that came to their mind. The answers were pretty good, but could be better.

But one day I got an idea: what if ChatGPT had more time to ‘think’ before answering (in case of AI it’s more like ‘thinking out loud’). I ran the experiment and it made the results so much better, but the cost of that thinking was too high to apply it to 10,000+ articles per day.

And then it came to me! (Machine Learning engineers will probably laugh at how obvious it is.) I could do two runs — first, a big and smart model could think really deeply about what makes the news significant and rank tens of thousands of articles (at a relatively high cost), producing a ‘ground truth’. And then it could ‘teach’ a smaller cheaper model to replicate its scores.

The smaller model probably doesn’t understand all the reasoning that goes into the scoring, but it doesn’t matter, as long as it is able to replicate the scores well enough. To give you some context: the previous approach of ‘instant scoring’ matched the ground truth 67% of the time. The new approach of ‘teaching’, raised the share of correct answers to 90%!

Second, I finally made it possible to read significant news by country.

This is literally the very first feature request I got:

It’s interesting that news by country follows the same bell-curve significance distribution! Here’s some examples:

I also added more features to a premium plan:

• email delivery — like this newsletter, but you decide what you get;

• topic blocking — you can hide news about anything: ongoing conflicts, politicians, even Taylor Swift.

• additional stories — unique stories that are only covered by one source;

To reflect the new features, I decided to increase the price of premium plan from $10/month to $15/month.

But I want to say thank you to you all — long-term readers who have supported the project all this time, even when it wasn't as polished. I’ll lock in the current price of $10/month for you, if you subscribe before December 7.

I recorded demos of all the new features. You can find them all (and possibly subscribe) here: News Minimalist premium.

If you have any questions — simply reply to this email. I’ll try to answer as soon as possible.

Have a great rest of the weekend and until next time!

Vadim