What is "ChatGPT-3" and which experts exactly regard LLMs as unimpressive?
MY feeling so far is more that laymen are less impressed of them than experts - because for people not from the field, the fact that they produce so much bullshit seems to trump all other aspects. Whereas I think for people who have kept track of AI developments a bit more, the fact just how coherent the bullshit is is still extremely impressive.
We've had random text generators for decades, starting with markov chains way back. Up until a few years ago, you were amazed if those things produced gramatically correct sentences - not even dreaming of producing any kind of coherent meaning.
In the potentially more difficult field of understanding text, there are a million small subfields, but nothing that could parse arbitrary english text and extract actionable, semantic meaning, in the face of idioms, figures of speech, complicated coreferences, etc.
Compare this to today, where you can give an LLM a written instruction in freeform and it will just execute that instruction - the answer may be glaringly incorrect, but it's exactly the kind of answer that fits your question.
In the field of understanding text, this is a progress which had seemed largely impossible before.