5 ms·
you know why LLMs repeat those patterns so much? because that's how real humans speak
by botusaurus 8mo ago
you know why LLMs repeat those patterns so much? because that's how real humans speak
- ndtimes 8mo ago[flagged]
- Starlevel004 8mo agoReal humans don't speak in LinkedIn Standard English
- chuckadams 8mo agoLinkedIn and its robotic tone existed long before generative AI. Know what's more annoying than AI posts? Seeing accusations of AI slop for every. last. god. damned. thing.
- IshKebab 8mo agoYes that's the point. LLMs pretty much speak LinkedInglish. That existed before LLMs, but only on LinkedIn. So if you see LinkedInglish on LinkedIn, it may or may not be an LLM. Outside of LinkedIn... probably an LLM. It is curious why LLMs love talking in LinkedInglish so much. I have no idea what the answer to that is but they do.
- roywiggins 8mo agoIt is at least thematically appropriate, of course a corporate-built language machine speaks like LinkedIn. The actual mechanism, I have no clue.
- swiftcoder 8mo ago"LinkedIn Standard English" is just the overly-enthusiastic marketing speak that all the wannabe CEOs/VCs used to spout. LLMs had to learn it somewhere
- kuschku 8mo agoHumans don't, but cocaine does speak "LinkedIn Standard English".
- cookiengineer 8mo ago> LinkedIn Standard English We need a dictionary like this :D
- macshome 8mo agoThe old Unsuck-it page comes pretty close. I’m not a huge fan of the newer page though. https://www.unsuck-it.com/classics https://www.unsuck-it.com/classics
- hiccuphippo 8mo agoReal humans write like that though. And LLMs are trained on text not speech. Maybe they should get trained on movie subtitles, but then movie characters also don't speak like real humans.
- shiroiuma 8mo agoMovie characters also don't speak like movie subtitles: the subtitles omit a lot of their speech.
- orbital-decay 8mo agoNo, they do it because they're mode-collapsed, use similar training algorithms (or even distillation on each other's outputs) and have a feedback loop based on scraping the web polluted with the outputs of previous gen models. This makes annoying patterns come and go in waves. It's pretty likely that in the next generation of models the "it's not just X, it's Y" pattern will disappear entirely, but another will annoy everyone. This is purely an artifact of training and has nothing to do with real human writing, which has much better variety.
- efilife 8mo agoYup, the first models always added "however it's important to note that..." at the end