Robot Pirate Ninja

So one of the sub-functions I’m using is a full-on summary and analysis tool. This is what generates the text in these boxes: (note: click the “more” button to see these on haikuthenews.io)

Each bit of summary text is a defined key:value pair in the original query to (currently) OpenAI.

This line here was the key to opening up the rest: response_format={ "type": "json_object" }

With that in place, you can define what each field you want to be called, and what’s going to populate it. This function also returns a Quality Score and Bias Score that can be seen at the top of each card. This was put at the top for a few reasons. The main one being that highly biased articles should be processed by humans only *after* they are aware of that bias. We don’t post-process information very well, but we have amazing pre-processing abilities.

So, we now have all the elements hanging out we need to do some solid charting.

At some point during dev today I decided to switch the little dropdown in Cursor.sh (apparently it’s cursor.com now) over to Claude 3.5 for a bit for this. Claude was quick, and got me moving fast. Eventually there was an issue and moved back to 4o.

Claude picked the “recharts” library (which I was not familiar with) and we start on the charts. Apologies I don’t have the works in progress, but we went through formatting, changed some axes, moved labels around, added a stacked horizontal bar chart (perfect for our bias direction metric)…and ended up with something that looks like this:

https://haikuthenews.io/institution/CNN

Or like this….

https://haikuthenews.io/institution/Fox%20News

And as we move over time, we should get a more accurate picture of various institutions, authors, and both…(porque no los dos?) and also I’ll probably add a compare function at some point.

I also added a bit to try and keep it from getting too busy, but with charts like this I didn’t feel like mocking up loads of data, so we’ll just have to watch and see how things play out.

Overall, I’m pretty happy with this, and will call it a day.

AI Ethics Sidenote: The summary bot is fully and completely capable of re-writing entire articles. Choosing how to use it is where ethics come in. I feel that what I’m doing here is additive to the overall value of the media artifacts being analyzed, and am fully open to further discussion.