Just read through and wondering if you could shed a bit more light onto how you’re decreasing latency? Also would like to learn more about the types of prompting you found to be effective.
Self hosting is huge - you don't have to wait in queues like everyone else, and can host in a region close to other servers. Similarly, just getting on the enterprise tiers of the services you use can also guarantee lower latency.
Not to be a de-cel, but what are the safety concerns around this? Can see this being great for pranking/fraud?
Yeah, all the safety concerns are very valid. Easiest immediate fixes that we're building into our system are 1) preventing people from scaling before doing KYC, 2) creating a content moderation filter on the AI's prompts