2 ms·
> If this instance of 'statistically close' corresponds very closely to what humans perceive as 'semantically close', we perceive this as parroting. Could it b
by visarga 4y ago
> If this instance of 'statistically close' corresponds very closely to what humans perceive as 'semantically close', we perceive this as parroting.
Could it be that with scaling the statistical model can become good enough for semantics? Quantity has a quality all its own / The map is the territory.
I believe we're seeing emergence from low level statistical patterns to semantics, especially in GPT-3, DALL-E and AlphaGo.