Reasoning & Learning
Explore several solution paths, learn from past runs, replay and fork runs, reason about cause and effect.
+++ 5 recipes +++
Tree of Thoughtsadvanced15 minA design question has many plausible answers. You want several branches explored and scored before committing to one.Self-Reflectionmedium10 minThe agent should notice its own mistakes — like an unknown ticker — and keep insights for next time.Optimize an Agentadvanced20 minYou have examples of good answers and want the agent’s instructions and few-shot demos improved from them.Time-Travel Debuggingadvanced15 minA multi-step run went somewhere unexpected. You want to replay it from a step, or fork it with a different instruction and diff the paths.Causal Reasoningadvanced20 minChurn went up after a redesign. You want to predict the effect of rolling it back and see the likely root causes.