Long-Term Memory
What's Worth Remembering and What Isn't
Users send dozens or even hundreds of messages to AI every day. Phrases like "Uh-huh," "OK," and "Haha" make up the majority; genuinely valuable preferences and facts may be just a handful. A memory system needs a gatekeeper.
Guess: Of These 10 Messages, Which Ones Should Be Remembered?
Think about which ones are worth storing in long-term memory, then click below to reveal
Gatekeeper Criteria
Memory Gatekeeper
Worth remembering
User preferences (likes/dislikes)
Important facts (birthday, city, occupation)
Long-term habits (early riser, vegetarian, uses Mac)
Explicitly stated requirement patterns
Important facts (birthday, city, occupation)
Long-term habits (early riser, vegetarian, uses Mac)
Explicitly stated requirement patterns
Not worth remembering
Filler words ("Uh-huh," "OK," "Haha")
One-off questions (What's the weather today?)
Pure small talk (no specific information content)
Outdated temporary information
One-off questions (What's the weather today?)
Pure small talk (no specific information content)
Outdated temporary information
If you remember everything, the memory store quickly turns into a landfill. A gatekeeper matters more than memory capacity. A good memory system focuses on knowing what to remember and what to forget—storage capacity is secondary.