Hacker News Reader

Just take an aggregate of every news article written in the top 100 newspapers.

That's high quality content, timestamped and about current events.

There is very little content on Twitter that compared in quality to one will written news article.

I would assume this is simply not enough data. You also have access to all public domain books and they usually make up a small fraction of the data used to train a model from scratch. For fine tuning a model, having a unique source of high quality data is probably valuable even if small.

There is information posted to X which the top 100 newspapers are not willing to / do not care to publish.

As we've learned in recent months, it goes both ways.

Reply on news.ycombinator.com