Synthesizers used to great effect
A while ago, I read a blog post by Mikey Dickerson that used the advent of synthesizers as a metaphor for LLMs. The author noted that he thought synthesizers created excellent reproductions of instruments, except the one he himself had studied. He quickly realized his fellow music students also felt this way about their own instruments. His conclusion was that non-experts are not qualified to evaluate the output of automated systems.
That's what I mean by the "synthesizer effect:'' If you know what you are doing, the machine can't fool you. If you don't, it very well might.
Trees, meet Forest
I’ve been thinking about this for weeks months literally years now, because it seems simultaneously “obviously correct” and “missing the forest for the trees”. It is true that synthesizers cannot currently replace a philharmonic orchestra with comparable fidelity to an experienced listener. It is also, perhaps more importantly, true that synthesizers have been massively empowering tools in the creation of music.
Both classical and electronic music can exist simultaneously. One does not devalue or invalidate the other. Nils Frahm is fucking sick.
I think we frequently choose the wrong frame when evaluating the potential impact of new technology. Instead of focusing on “what am I currently doing that I can replace with this tool”, we may be better off with “what am I able to do now that was previously impractical.”
Everyone is like AI can't make art but it can sure as hell take that idea for a little MIDI sequencer based on nested lists and help you get the Python done in like an hour with comments + tests, then translate that to C++ so you can start to build a VCV Rack plugin, for about 70 cents in API fees. @ftrain.bsky.social
Typing is not the bottleneck
Another way this same concept has surfaced for me is in the phrase "Typing is not the bottleneck." Again, I think this is true, and missing the point.
I'd had a design sitting in my head for years. Never could motivate myself to do it just because of how tedious it is. With Opus 4.5 I got it done in one day
We experience a "bottleneck" based on some course of action we have decided to pursue. The things which are not bottlenecks are other constraints that exist, but are not directly limiting our productivity or progress. When we choose projects to take on, or problems to solve, we do so within a complex "possibility space", bounded by a large number of potential constraints. If we loosen one of those constraints significantly, the shape of the possibility space can change, and the decisions we make within it can also change.
how do you quantify this productivity gain? it's not a straightforward speedup because I was simply not writing vale rules before, so I don't know how long these things would have taken. more importantly, even if I had spent the time, just counting the saved minutes doesn't account for the tedium
Again, to me the more interesting question is not "does this help me do the things I'm currently doing cheaper/better/faster", but rather "what new things might I choose to do now that I wouldn't have considered otherwise".