Share this postEmerging Market SkepticEM Fund Stock Picks & Country Commentaries (February 16, 2025)Copy linkFacebookEmailNotesMoreEM Fund Stock PicksEM Fund Stock Picks & Country Commentaries (February 16, 2025)AI experiments to improve these posts + Substack tech glitch. DeepSeek deep dives, searching for black swans, South Africa in a cyclical uptrend, Argentina turnaround, dollar "Mar-a-Lago Accord?" etcFeb 16, 2025∙ Paid1Share this postEmerging Market SkepticEM Fund Stock Picks & Country Commentaries (February 16, 2025)Copy linkFacebookEmailNotesMoreShare

A quick note about some experiments with AI and improvement to these posts: I’ve been playing around with DeepSeek (when the server is not giving a busy message - which is a good portion of the time…) and another Chinese AI service that has not been hyped as much (and I won’t mention by name as I don’t want it to start giving a “server is busy” message!).

Both services can be used to extract stock names or other key information from fund updates or my Monday links posts and organize them into tables (albeit DeepSeek, when the server is not busy, produces a less reliable table output with missing stocks/articles…). For example and from my Monday post using the other Chinese AI service - my request and the truncated output (I only screenshotted the top of the table…):

We subsequent requests, I was able to get them to include the embedded links into the previously produced table to make the article headline clickable to the article. However, I could not get them to embed the links for the stock names and tickers (as for some reason, the AI kept searching the internet and including Google, etc. links…) plus the AI would run out of steam (or server capacity/bandwidth. etc.) before completing the task with all stocks and articles from my Monday post.

In addition, I used the AI to extract/list stock names from some several page fund annual reports or transcripts covered later in this post. However… there are still AI teething issues e.g.:

• The AI gets confused if you don’t start fresh conversations b/w completely different requests - especially when uploading files - and you often need to make follow-up requests…

• Some words don’t register with the AI e.g. asking for which stocks on the list were detractors or detracted (the most common word funds used to describe weak performers) seem to miss finding stocks that are described as having “weakened,” etc. This means everything still needs to be rechecked - something that can also be time consuming (just like outsourcing work to a third party in a foreign country, etc.)

Nevertheless, the AI’s ability to extract some basic data/info and organize it into tables or lists is very useful/helpful.

Note that I just listened to an alt-finance type of podcast interview where the point was made that AI is not necessarily going to cost people their jobs, but rather the people who use AI (or know how to use it) will take the jobs of those who don’t know how to use it or incorporate into their work…

However… my next problem is with Substack’s continued inability to allow writers to create tables from within the platform as they continue to be focused on becoming Twitter/Tik Tok/Spotify etc. with a blogging platform thrown in.

Substack does have an integration with a useful table/chart/graph generation service called Datawrapper; but the table must be created in and can only be edited within Datawrapper plus it appears links cannot be embedded into the tables. In other words, not writer friendly. I also don’t like the idea of relying on a third party service e.g. Tweets still cannot be embedded into Substack due to them pissing off Elon Musk with their Notes clone…

With that in mind: I have further cleaned up and simplified our continuously updated post containing all funds (who’s summarized changes then get emailed out on Sunday) - for example:

And please let me know if the following emojis cannot be read by any particular browsers or phone models: