Context is temporary working material, not permanent knowledge
What you put in a conversation is not stored knowledge, it is scratch paper for that exchange.
People get this backwards in both directions. Some assume the model has learned what they told it and will know it tomorrow, which it will not. Others assume something typed into a chat window disappears, when it may be retained by the provider, logged, or reviewed. Neither assumption is safe to make without checking what the specific service actually does.
More on AI and LLMs
- An LLM predicts plausible continuations, not verified truthIt only checks the shape
- Retrieval adds documents, not guaranteed correctnessThe filter slot is empty
- Temperature changes variation, not factualityThe dial only sets the spread
- System prompts are instructions, not a security boundaryA sign, with no fence
- LLM output is untrusted input downstreamIt comes in round the back
- Fine-tuning changes behaviour, not every limitationNew type, same roller