New CMO + Enhanced Scientific Advisory Board, page-17

my understanding is that a SNN and CNN are quite similar with the key difference being an SNN only runs the full model on non 0 inputs where the CNN runs the full model on every input. the CNN2SNN converter was essentially adding a preprocess to determine non zero inputs to run the full CNN model on. it wasn't actually changing the underlying network structure. a TENN is a completely different network structure and so the conversion can't be applied in the same way I don't think. in terms of model perameters GPT-4 is estimated at 1.76 trillion perameters where the TENNS model in the paper has 196 thousand