Safer Centaurs
I am exploring how to use AI better. I find for example that AI blurs concepts. New ideas tend to be reduced down to more mainstream versions, unless you keep pushing at the novel aspects.
AI excels at software. I built the code on this site with a great deal of assistance from AI. In the blog posts here I explain exactly how I did this, what worked and what didn't.
-
The '📣 announcements' are of code that has got to the stage I intended for that exploration. It may not be polished, but it is usable.
-
The '⚡️ work in progress' are ideas I am working on, with steps along the way. These are 'spike solutions', code that shows an idea from a particular angle. These are some of the most informative for what AI gets right and what AI gets wrong.
As well as working on the code I also discuss various topics with LLMs: Mars colonization, graphics that morph, machine-tools design. I talk about some of these discussions in these posts.
What AI can do
The LLMs are great at surfacing information. They work as 'a better web search'. Their logical reasoning about what they find is currently so-so. So this is a good opportunity for human-ai dyads, the AI to search and bring things together, the human to guide the process and wrap some better reasoning around what they find.
A year or so ago it was tenable to say 'AI is not creative'. Nowadays it no longer is. AI clearly can assemble original new combinations of things, and blend those combinations together nicely. Defining what AI can do well and what it can't, what kind of creativity it is capable of, has now become a more nuanced thing.