Quality & taste

I’ve been using LLMs as a bicycle for my mind, learning things at an incredible rate, writing code in languages I’ve never used, and trying out new ideas. It still feels like magic. It’s reshaping how humans work and think. We should all pay close attention.

But here’s the thing, I don’t think LLMs will replace the meat-computers in our heads anytime soon. I think this is because they don’t have the language to understand quality, nor do they have taste. Not now, probably not ever. They’re weird tools full of ghosts. And like most tools, they need a human operator behind them. Someone with judgment, taste, and a vision of what they’re trying to do (even if they can’t say why). Without vision, all you get is nonsense.

But vision is hard. Direction is hard. You don’t always know what you want, or you can’t express it well, or you change your mind halfway through. An LLM can’t fix that. It just reflects whatever you give it. It doesn’t care.

So what is quality? For me, it’s a feeling more than a definition. It shows up as a side effect of when someone cares… when they pay attention to the details without loosing sight of the whole, when they make bold and intentional choices, when they put their heart in their work. You can feel when something has quality, and other people can feel it too. Humans are so good at this that we often forget that it’s a thing we can intuitively sense.

Taste is something different. It’s both intensely personal and contextual. It comes from experience, mostly from trying things and reflecting on them. I’ve made a zillion mistakes. I don’t write them all down, but they stick with me. Over time, that kind of learning builds a sort of internal compass. It’s not always conscious, but it helps you make better decisions. I think that’s what taste is. When you say “that’s not how I would do it,” that’s your taste talking. You have an opinion. You know what fits and what doesn’t, even if you can’t explain why.

An LLM’s taste starts and stops with the data it’s trained on. It carries no strongly held opinions or hills to die on. An LLM will never know what human experience means. Not unless you describe it to them somehow. And even then, you’re just passing along the gist of it, never the real thing. Maybe that will change. Maybe we’ll get better at designing ways for these systems to “get it.” But for now, us meat-computers still have the edge. Because we feel ways about stuff.

And feeling, at least for now, is still the difference.