How do i stop the charges? Has the responses got better in new LLM models? i promise you though i'm super chill, so don't be afraid to reach out to me if you're interested even the slightest.
Now, reap what you sow. First post on reddit.
This really only becomes obvious once you're watching the same clip repeat 10+ times across testing which I do well over that as i'm testing haha, but would love to hear what others have figured out! I feel like heβs only given advice only to cis guys. DM's open.
The only thing I didn't like was when Joey had a crush on Rachel, and Phoebe found out about it and felt rejected. But we can give money too. NO gore, violence, calls for violence, porn, and just anything overtly nasty that serves no other purpose than to be nasty. **The characters:** - π€ Zael β 27, vampire, dry humor, replies with "k" but you still want his attention - β‘ Draven β 32, self-made CEO, every word costs money, warm only in flashes - πΉ Ragini β 35, Ahmedabad woman living alone in Mumbai, Hinglish, always 3 steps ahead - πΈ Zara β 24, Ukrainian model in Mumbai, scattered and warm, accidentally wise **What actually worked:** - Character prompts with concrete DO/DON'T examples beat generic personality descriptions - "FIRST RULE: Always respond to EXACTLY what they said" in every prompt β eliminated generic responses - Polling over webhooks β zero dropped messages - SQLite over in-memory β conversations survive restarts - Groq 70B over smaller models β huge quality jump for character consistency **What didn't work:** - Free OpenRouter models alone β too unreliable, rate limits hit constantly - Generic prompts β AI defaults to assistant mode without very specific instructions - Long context windows β keeping last 10 messages is enough **Current status:** 5 users (just launched, not promoted yet) **Next up:** Echo β a group presence bot that sits in Telegram groups and creates curiosity, funneling users to the main bot.