"proven wrong" is a strong word, but this is not the direction we seem to be going with AI at the moment.

If you're trying to get an intelligent system by mostly mimicking human behaviour (+ some generalisation and finetuning) then it seems somewhat obvious you're going to hang around human level and amplify the differences there. AI systems trained with less reliance on human data seem to blow past human performance pretty easily. (E.g. the Go version of AlphaZero re-writing a lot of how Go is played at a high level and throwing out things humans long considered optimal.)

the slow intelligence growth rates are just because we're using llms to make ai, the tiny gap theory wouldnt get disproved unless someone showed brainlike agi taking a long time to cross the gap

It's taken five years to get from primary school maths to graduate maths, hardly tiny