MathiasTCK.bsky.socialNo. #LLMsDoNotUnderstand , that's not what they do or how they work.
But I think I get it. LLMs seem alright at generating text that seems to indicate understanding certain material better than the average person on lots of topics, especially topics the average person won't know much about.
But like, only when the words are coming out of a person capable of understanding do they really indicate understanding.
Like it's obvious the TV does not understand the news that gets played on it. The TV is showing evidence the newscaster understands, or their writers understand a topic.
And things like context, tone of voice, diction, further convey understanding.
LLMs producing content is evidence their data was sourced from people that understood things, and the LLM is OK at generating related text. But only as good as it's training data and model allow.
#AbominableIntelligence