Were there any concerns about GPT-3's latency? It looks like it takes a long time for online use cases.