well that’s tremendously exciting that you’ve cracked a debate that experts in the field are still debating. and the fact you’re so certain about it means you must have some really solid work
look forward to reading it once it’s through peer review
why would someone say such transparently uninformed things like this with such certainty
i respect if somebody takes the position that they don’t agree. fine. but falsely claiming that there is not academic debate over something purely because you’re unaware of it is just weird and bullshitty
Not true; they have zero conceptual frame, only local linguistic facility. If AI *understood*, it would know that a human has five distinct fingers, instead of arbitrary fingerlike glop. The number of "r"s in "strawberry" would be easy. But no; an LLM is just autocomplete with a trillion parameters.
if you want to talk about why stable diffusion models - which are not LLMs - have glitches like this, that’s an interesting side discussion . the Rs in strawberry case is about LLMs and is fascinating, and also solved
none of it has much bearing on how concepts are represented, except rhetorically
Comments
look forward to reading it once it’s through peer review
i respect if somebody takes the position that they don’t agree. fine. but falsely claiming that there is not academic debate over something purely because you’re unaware of it is just weird and bullshitty
none of it has much bearing on how concepts are represented, except rhetorically