Subject: Re: Anybody read BRK-B.com?
Hey everyone,

thanks for the heads-up!

It's my page, so I presume it's ok I answer - and I'm not a bot :)

TLDR:
Initial Article generation is done with the help of gpt-4 based on certain information / resources, then manually edited.
For shareholders studying the annual report themselves, checking out BHE's, BNSF's and other subsidiary pages regularly there is probably little value added.
For other people interested in Berkshire there might be something they did not know already.


Long Version:
For this site I am using an approach to outline my line of thought based on different sources (like the annual report). Afterwards gpt-4 is formulating the initial text.

That gives me a first version of an article and I go over it, add some further information here and there to (hopefully) improve the article. Note: I have outlined usage of AI on the about page.

Why this approach? It enables me to get much more resources into one article. Like general statements of the economy, how the insurance industry is doing, information like these.

This is good as I simply could not do this in this exhaustive manner and with such elaborate English (not native speaker here). And it is, to my mind, much more interesting than the "Berkshire has moved 1.37% today, why" or "What could you have won if you would have invested in Berkshire in 1987" kind of articles.

I was simply looking for a site on Berkshire focusing more on the subsidiaries and not the investment portfolio. And I think there is not much out there. Hence I might be able to provide some value with my approach.

The downside is of course that things can go wrong twice: my understanding of something (thought wrong) or gpt-4 getting my correct thought wrong. (I think the article on the yen bond strategy is one example for me outlining the idea wrong and gpt then just outlining a general argumentation)
And, of course, articles tend to be longer than needed.

Note that I am trying to provide the direct references to the used sources in the text. That's a bit more effort but it is easy to verify certain facts. So even if I'm wrong everyone can find their own interpretation.

Generally it's a question of effort vs. result. There is no Pulitzer Prize waiting for me, I want to just provide some (hopefully) useful information, I am not a journalist. I believe the quality of writing enabled by my approach is much higher than if I would write the articles myself. This is a general AI phenomenon.

On the AI images thing: I don't have access to stock fotos and I simply like it. It's (mostly) some comic images transporting some basic information what to expect in the article. Nothing complicated, nothing offensive.

These are my 2 cents, I hope they were somewhat clear.

If you have any suggestions on how to improve the quality of the content, just let me know. I will be happy to adapt.

Take care

Robert