Remix.run Logo
anilgulecha 2 days ago

The premise of Nolan's position is Claude's language tic should not rub off on him, vs it's ok for Tolstoy's to do so.

I disagree. I think words are powerful, and when a word will be understood exactly as it's stated, then the value of the word is present. So while he might bemoan "load-bearing" (so would many of it's readers), that word in the right place will exactly communicate what he'd like. That enough reason to "earn it's keep" in prose.

paimapi a day ago | parent | next [-]

it's not that load-bearing isn't doing the, well, load-bearing semantic work but it's both the lack of variability and lack of nuance hat makes it sound cheap. word choice is powerful - and in the same way, it can be nuanced

for eg. consider the synonyms for 'load-bearing'. if you want construction motifs: keystone, linchpin, cornerstone, foundational, structural. if you want earthy: bedrock, root. if you want ten-dollar words: axiomatic, sine qua non. if you just want to-the-point: vital, key, tenet, crucial

each of those words contain nuance. vital is used a lot in medicine (vital signs) so health, the heart, etc are evoked. pair it with specific usage like "this point is vital, the argument would die without it" and it rings different than "this point is load-bearing, the argument would be refuted without it"

you'd need to exist as a purely logical, probabilistic semantic parsing machine with no innate feeling or cultural attunement to prefer the lack of color and nuance. language as fully utilitarian with no artfulness is dreadful - if all I had to read all day was a self-important tech manual, I too would be frustrated and want for better craft in output prose

anilgulecha 18 hours ago | parent [-]

I didn't say I'd prefer it. I said that's the only word that presented itself to the author, it provided value, and there's no reason to feel bad about it.

fjcururuvy7 a day ago | parent | prev [-]

The fact Anthropic, OpenAI and Google think they can fingerprint model output with no fidelity loss tells me otherwise.

As does caveman.

goodmythical a day ago | parent [-]

Why? We can also fingerprint human authors with high degrees of certainty if given large enough samples.

We've been doing so since at least before the media excitement over Robert Galbraith being outed via software analysis as J.K. Rowling.