Title text:
You say no human would reply to a forum thread about Tom Bombadil by writing and editing hundreds of words of text, complete with formatting, fancy punctuation, and two separate uses of the word 'delve'. Unfortunately for both of us, you are wrong.
Transcript:
Transcript will show once it’s been added to explainxkcd.com
Source: https://xkcd.com/3126/
I got accused of being an LLM for the first time just a few days ago. Was pretty funny.
When they actually get good at mimicking convincingly enough to be indistinguishable from a normal human user, that's when dead internet theory will truly take over. This could've already happened, but I've seen enough stupid shit vomited by LLMs to know it probably hasn't happened yet. Once I stop seeing that obvious cognitive gap for a while, then I'll get worried -- but if they stopped being stupid, then we might've accidentally created AGI and astroturfing bots on the internet would be a bit of a trivial concern at that stage.
*in English.
They still suck at many other languages, like Finnish.
They're still garbage in finnish. Even if they manage to make grammatically correct sentences, it sounds like incredibly awkward word salad, often there isn't even real meaning in the sentences. Non-human translations have struggled from the beginning and I'm not surprised even the LLMs aren't doing that well.
Kuulostaa siis ihan paskalta ja on edelleen tosi helppoa tunnistaa. Ei akateemikot väännä sellaista tuubaa omin käsin, eikä kukaan normitason juntti taas kirjoita niin jäykästi. Tai ok varmaan oman elämänsä jarisarasvuot yrittää suoltaa samankuuloista tyhjää konsulttiroskaa, mutta olettaisin että ne on juuri sitä porukkaa jotka käyttää tekoälyä muutenkin
They've mastered Danish, though. Or at least the native speakers can't tell the difference.
https://youtu.be/s-mOy8VUEBk
None of the other people in the Nordics are surprised that the native speakers can't tell the difference. We can't either after all