Writing this to see how it has aged in a few years.
I held back on using AI for a few years, especially running it on the cloud (meaning a datacenter which is wasting good water). Played around a bit with local models on an Nvidia 3080 but the results were meh.
In the end, I decided to go and experiment a bit more with it while it lasts.
Pretty sure that in a few years the AI as we know it now will go the way of the dodo, just not certain how. But if we look at the past, the giants of yesteryear are in quite a few cases the dodo’s of today (CompuServe anyone) or they pivot into another reality completely (Nokia anyone? ).
Maybe in a few years there will be no more AI, or perhaps only local-run AI’s. Perhaps on a small chip card. Who knows.
I decided on using Claude for a year, and I have to admit, it impressed me. Creating a program in python + flet, wiring schemata’s…. it can do it all.
My conclusion for Claude: as long as you know your stuff and get detailed enough, you can get it to create ‘stuff’ for you that would take ages for you otherwise. Just don’t be afraid to say that it must stop focusing on the wrong bit, and tell it to look at other alternatives. That also helps you save your credits.
At the same time, it scared me how easily I would think of ‘Ask Claude’ for trivial stuff, so I try to stay away from it for things I can go and search for myself (and even there the first response you get is usually from AI).
We’ll see how it goes.