|
|
|
|
|
by tony_cannistra
13 days ago
|
|
TFA raises several points I hadn't heard before. Here's one. The artifacts we produce define our culture. If we allow LLMs to produce our work, our culture represents that which they understand. But LLMs can't be trained on the entirety of the human experience. "How dare we abandon so many countless traditions and people and ideas and nuances, simply because they are underrepresented in the training data?" I just feel like that's not performative in the way you're insinuating. That feels like a take to me. Respectful engagement is what I expected here. |
|
LLMs are a tool. The brush is a tool, the pencil is a tool. The way we write already is defined by tools. This very comment is remarkably defined by its medium, I can't control its font size! Or the font at all, for that matter, nor the color, hell I can't even control how and where this comment is read! This means I'm generally going to write it in English, a whole different language from my native one! How is that for "defining"???
None of this is new, it's all dictated by human experience in the very same way that the output of a LLM gets dictated by human experience. That's the way memes (as in, anthropology: https://en.wikipedia.org/wiki/Meme?) work.
LLMs don't understand, they just output based on some parameters. We've been here before.