I got voice calls working on the Surface Book. Video, not so much. Turns out Microsoft didn't write any linux drivers for their custom camera hardware. Grok happily wrote one for me. The funny thing is that it worked super hard on it until it actually took a picture which came out garbled. Since then it has been depressed or something. Just giving short answers to my questions and doing no work. I finally got it to rewrite the kernel module to sweep through all possible settings for the camera. We'll see how that goes.

Whoa! I just visited the #amethyst settings for the first time in a while. Vitor's been busy. I'll have to dig into these.

The supposedly imminent AI apocalypse is very real and very overblown. I don't mean that there aren't real risks, but while the particulars are unique, humans aren't going to end up in a particularly new situation. Intelligence, artificial or not is merely a way to possibilities into actions. Let's assume that AI far surpasses our own intelligence as does its agency to act in the world. We'd end up living among inscrutable forces acting beyond our comprehension or power to resist. Sound familiar? This was our ancestors living beneath the notices of the capricious gods of the weather stalking beasts. It was the serfs living at the whims and generosity of their feudal lords. It is us hapless before the required paperwork, phone-trees, government workers just following the rules. If we can survive volcanos and the DMV we will probably survive ai.

Something good is going to happen, something small, almost unnoticeable, but it will slowly tilt the world when you do it.

I am also toying with a genetic algorithm approach to growing larger AIs by combining smaller parents with a shared ancestry. It is a way to allow us little people to collaborate on training AI by each of us doing relatively cheap fine-tunes of selected ancestors. For instance you may choose to combine an expert on gardening with an expert on orchids and supplement with some specific orchid gardening curriculum to give yourself a new model that excels in that area. The ancestors do have to be fairly closely related or you will get gibberish but I am hoping to get it to be robust accross slightly more distant relations. Lots of other ideas I want to try, but this should allow testing just about anything.

I am tasking grok bot to scan articles in regional languages around the world to detect unusual messaging. Hopefully I can be a bit more alert to what someone somewhere really wants me to believe. It will probably take some fine-tuning.

From my wires crossed 5 year old. "The Vikings won in straight sets!" Silly girl doesn't realize that only counts if they can score 3 baskets in the tenth frame without giving up a run.