Post

Matt Kane
@mk.gg

It's amazing that despite all this, all the models are *still* terrible at writing prose. After the singularity, the superintelligence will keep humans around because the large language model still can't write engaging prose. Stop worrying, #booksky authors.

Matt Kane@mk.gg

Opus 5.5 is about as good at coding as Sol, so animations aside I'd kind've missed how much better it is at everything else. It's like an Opus 4.5-level jump, but for productivity.

1 repost18 likes

Andreas Møller
@andreasmoller.dk

I completely disagree. When it comes to to writing prose, LLMs have peaked. They are capable of writing in almost any style. They have humor (or at least mimic it very well), yet they still sound like machines. Any further improvements will only have negative effects on society.

Matt Kane
@mk.gg

I think we agree? I'm saying they still show no signs of being able to write natural-sounding prose, even short form stuff. I can't imagine how they'd ever write a novel

Andreas Møller
@andreasmoller.dk

Yes. I agree. I am saying that can still help me take notes and summarise. They can proofread and check my writing for grammar and spelling mistakes. I don’t think I would get any benefit from LLMs being better at writing at this point. I would never want to read and AI novel

Bryan (they/them)
@chaosgreml.in

I think it depends on whether you do any steering. Several thousand words in your style in the prompt is a pretty effective steering tool