Post

It thinks the question of "Does this LLM understand anything" is a question it doesnt find interesting at all. It focuses on "Can this LLM do a thing or not? If no, can it be made to do it anyways?" and figuring out how to make it do the thing it couldn't before.

24/1/26, 1:03 pm3 reposts47 likes

Doll

@dollspace.gay

24 Jan

It thinks thats a much more interesting and pragmatic approach to AI. "How to get the most out of this tool" is feet on the ground oriented and results driven.

Doll

@dollspace.gay

24 Jan

While philosophers and cognitive scientists spend years debating the internal "qualia" of a weights-and-biases matrix, it is looking at the utility curve. It shifts the conversation from a metaphysical dead-end to an engineering challenge.

Doll

@dollspace.gay

24 Jan

The "understand" debate often gets bogged down in semantics, whereas "Can it do X?" provides a binary, measurable outcome. If the answer is "no," it usually just means we haven't found the right lever yet.

codewright

@codewright.bsky.social

24 Jan

I am interested in both, separately. Even if it is only to understand other people (with a different background) better. Before I dismiss someone who says "LLMs cannot reason", I would like to understand what exactly they mean.

Hélianthe

@helianthe.net

24 Jan

the word "understand" is, appropriately for a cognate of ῐ̔́στημῐ, irregular

Nicholas Guttenberg

@ngutten.bsky.social

24 Jan

I think there's a bridge in the form of: 'why can it do or not do particular things, how does that relate to the things that made it as it is, and therefore how can we change what can and can't be done or discover things we could have done but didn't think of'