"which they have never actually used to play Tears of the Kingdom (alt text)"

I feel seen. I COULD be playing TotK right now, but that's such a heavy lift when Pokémon Puzzle League is right there on the N64 VC in all its pixellated glory!

(Well, not necessarily *easy*. It took me hours to write this thread!)

Part of the problem is that most writing and talking is done by people who write and talk a lot — so the discourse is disproportionately focused on use cases where LLMs are helpful for that group. It's easy for the rest of us to dismiss those use cases as just more magical thinking about AI.

Someone who finds LLMs useless in one scenario may find it useful in another. For example, I struggle to get value from conversational interfaces, but I do find it useful to have a local LLM generate a first draft when I write descriptive image text.

When writing an English essay, a Greek-English dictionary is useful if your first language is Greek and useless if your first language is English. Similarly, an LLM that writes code to a description is useful if you find it easier to write descriptions, and useless if it's easier to write code!

I think of LLMs as trying to translate between "natural" language and other forms of expression — other natural language (summaries, writing styles), structured language (search queries, computer code), or something else (images).

Like any translator, that can be very useful or not at all!

I have a theory that people's opinions about LLMs depend on how their brain works.

People good at verbalizing: LLMs are so useful! I can get so much done just by having a conversation!

People (like me) bad at verbalizing: LLMs aren't useful! I have to have a conversation to get anything done!