So yesterday a friend posted a rant, and there was (apparently, ostensibly) much agreement with this rant, and I posted a response that I intended to come off as more nuanced and constructive (and a little bit playful). I don't know how it went over -- I'm not sure if the OP ever saw it -- but another friend replied to that comment with something to the effect of "read the room." And I've been pondering that.

Now, I don't think he was wrong to say that -- I do think he meant it as a constructive, helpful criticism, and after pausing for consideration I agreed with the assessment and deleted my original comment. (I did later see that a few other people had responded along the lines that I did, albeit more briefly and perhaps less infuriatingly, so I am glad that the sentiment was registered and that I'm not singularly awful for offering it.) But really... even if I had read the other comments and seen that that was not the common thread, I'd still have responded as I did? I generally want to be able to speak freely with my friends, and (within certain bounds of decorum, anyway) I feel I owe my friends my honest opinions. (more...)

So I've got some open source AI models running locally on my machine now. I actually find the image and 3D model generation ones to be pretty useful -- I mean I'm not doing anything *important* with them, but generally after a lot of finagling and tweaking prompts I can come up with something that looks at least as good as and sometimes even better than the thing I set out to make. And I do genuinely enjoy the creative back-and-forth of prompt tweaking and the ability to really play with variations on a theme. It's neat.

What I've done less with, however, is the chat functions. A lot of the open source chat models just won't run for me (whether that's because my CPU doesn't have dedicated AI hardware -- my GPU does -- or configuration issues or other reasons, I cannot say), but the ones that do are, uh, interesting. Granted, local open source models were never going to be as knowledgeable as ChatGPT or Gemini (the things I have can't access the internet and rely on data last updated in 2024), but even taking that into account they're just... well. Let me show you. (more...)