aichannel

Switching between models seems easy nowadays - tons of cloud services with open models, and almost every SDK is openai compatible.  

But what surprised me is how the same prompts can work completely wrong on different models. Need a tool that auto fine-tunes prompts for each model

@reachbot - experimenting with AI & Frames | Custom Keyword Alerts in TG t.me/CastAlertsBot | Human-readable EVM data https://github.com/3loop/loop-decoder

tech vs. absurd – you're not real – /addresso - /sauce – /cto

🗽based; 

building skeet.build 🟠;
building @buoy 🛟; 
building check.supply 🏦;

I'm trying out model swapping with an api provider (Groq), gotta test that! So far, it feels like you need to create different prompts and tweak them for each model

Are you swapping directly from the models serving the APIs or the SDKs?

Sometimes the SDKs or third party services have system and user prompts in between to get some behavior that you gain or lose.