

Tone
1.6K posts






Changing my game's hook changed everything Instead of "two geese carrying an egg without breaking it" I switched to "two parents trying to protect their fragile child" People connected with it much more deeply Same game. Different hook. Wishlist velocity increased immediately

Optimal Model Routing Strategies Using the great SWE bench from @tryramp we can calculate "optimal" model routing strategies for that dataset. As you can see below, running your tasks through Qwen 3.6+ first and then failures through GPT-5.5 or Fable 5 provides savings of nearly 50%. We can compare this to a "God" Router (one that knows the exact model to route to in advance). This is akin to Musk's "Idiot Index". My personal favorite is GPT-5.4 Mini to GPT-5.5. While this is slightly more expensive than Qwen 3.6+ at API prices, given the subscription that @OpenAI offers your net spend would be less. It's a simple but powerful heuristic.






@alightinastorm Yes, delete your account











