Unverified50% confidenceFactExact time
Large language models themselves are stateless, and each request is an independent computation; so-called 'remembering' usually relies on two mechanisms: one is concatenating historical conversation into the current prompt (the context window), and the other is retrieval-augmented generation (RAG) through external storage (such as vector databases or file systems).
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn engineering implementation, the persistent memory layer typically exists independently of the large language model itself, as an external service or database80% similarUnverifiedAnother technical approach for large language models to implement memory is to use external vector databases (such as Pinecone or Weaviate) for semantic retrieval of learning records and dynamic extraction of relevant segments.76% similarUnverifiedLarge language models are essentially stateless functions, and a model's memory depends entirely on the content within the context window.75% similarUnverifiedThe "forgetfulness" of large language models stems from their stateless inference architecture: each time the API is called, the model only processes the token sequence passed in the current request and retains no internal state across requests.75% similarUnverifiedLarge language models do not require persistent storage of user data and can complete real-time inference solely based on natural language input from a single conversation.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1016684API
curl https://kongchang.com/api/v1/knowledge/claims/1016684MCP
get_claim(id=1016684)