Conversation
Replying to
➡️ 1/6 My original intent was to see how far you could push GPT-3 to generate a recipe based on just one priming example being given.
1
1
➡️ 2/6 It will twist & bend parts of the original priming recipe to create a new one. The more your ingredients make sense for a “real recipe”, the less it will try to fallback to the priming recipe, and the more confident it will be to generate something real & legit.
1
1
➡️ 3/6 While it's probably not always correct... well actually I can’t judge since I’m not the best cook myself 😇. Anyway, it does give instructions on how to prepare (cut/chop/...) each ingredient while also giving clear directions on combining them. That’s very neat!
1
➡️ 4/6 As I mentioned before, this demo has only been given 1 example recipe. You could probably enhance the outcomes by priming it with multiple ways of cooking & baking so it has more variety to choose from.
1
➡️ 5/6 Lastly, the most interesting part is that I've only primed it with a recipe of a basic salad. Yet this was enough to generate detailed instructions for a complete lasagna. 😄
1
2
➡️ 6/6 I plan on doing more demos & experiments like this, so if you like these demos, make sure to follow me for more fooling around with the API 👋🏾
1