in terms of LLMs that we're using for vibe coding? those models are only using knowledge given to them by other humans.
Discussion
Not sure about all models, but Claude 3.5 was trained by Claude 3.0 to make rational arguments. Reason + randomness = creativity
They can produce creative thought (given the “right” temperature) they just can’t tell what’s utter nonsense and what’s genius, which has always been the case actually.
This is where a sophisticated sense of taste comes in.
nevent1qqs9f4wrnadqy7n54cxzke8p0h0z7g7ze5e886ktxgzqvmv46k6p7ssppamhxue69uhkztnwdaejumr0dsa9ux7w