New: Intelligent Flashcards spaced-repetition decks for every topic, built to make things actually stick.
With Reflexion, how does the model carry the lesson from a failed attempt into its next try?
It permanently rewrites the system prompt so every future user shares the same correction
It updates the model weights using the reward signal so the failure is baked into the network
It writes a short verbal self-reflection and keeps it in its memory as extra context for the next attempt