> I’m encouraged by people who talk about their AI setups using the same words that musicians use to talk about their guitar amps or pedalboards. This is a good sign. Obsessing over gear - not over purchasing gear exactly, but anything that feels like “how do I get a particular kind of crunch out of this preamp” is a great sign for the medium because it’s a format where you look at your own output serially over time and develop a style that you actually want.
Hmm, I tend to dismiss this idea mainly because I think it'll soon become unnecessary and also belies why most people have "custom setups". My feeling is most people are using LLMs to achieve concrete goals: How to do some basic woodworking, bake bread, get a condensed version of a college course on nuclear physics, or write code to accomplish a task.
Right now specialized setups and finely-tuned models might make sense for bridging the gap between "almost there" and "good enough", but the overall trend seems to be moving toward general-purpose LLMs becoming “good enough” to handle most of these tasks. Over time, the gap between a highly specialized model and a general-purpose one seems to
shrink for the level of expertise most people are looking for.
No doubt there may still be some customization for more novel creative applications (and the author even touches on one I expect to see-emulating the dreamlike aesthetic of early generative AI). But novel creativity is a small minority of "My AI Setup" type articles that I see at the moment.
Hmm, I tend to dismiss this idea mainly because I think it'll soon become unnecessary and also belies why most people have "custom setups". My feeling is most people are using LLMs to achieve concrete goals: How to do some basic woodworking, bake bread, get a condensed version of a college course on nuclear physics, or write code to accomplish a task.
Right now specialized setups and finely-tuned models might make sense for bridging the gap between "almost there" and "good enough", but the overall trend seems to be moving toward general-purpose LLMs becoming “good enough” to handle most of these tasks. Over time, the gap between a highly specialized model and a general-purpose one seems to shrink for the level of expertise most people are looking for.
No doubt there may still be some customization for more novel creative applications (and the author even touches on one I expect to see-emulating the dreamlike aesthetic of early generative AI). But novel creativity is a small minority of "My AI Setup" type articles that I see at the moment.