Three memories people always confuse

"It remembers me." The phrase covers three mechanisms that have nothing to do with each other, and confusing them leads to two opposite mistakes: believing it retained everything, or believing it kept nothing.

1. The context: what it has in front of it

The current conversation, the attached files, your instructions. All of this is sent back to the model with every message. When you open a new conversation, this context starts from zero. The Beginner level showed it: this is a working memory, limited, and the middle of a long exchange is read less well than the beginning and the end.

2. Persistent memory: notes the software keeps about you

Most major assistants now offer a "memory" that survives from one conversation to the next. In practice, it is not the model that has learned anything: it is the software around it that records notes ("training for a marathon", "works in accounting") and quietly pastes them back at the top of each new conversation. Some go further and also draw on your past conversations when they see fit.

Direct consequences: these notes take up space in the context, they can be wrong or out of date, and above all they can be read, corrected and deleted, in the settings.

3. Training: what it knew before you

The model's own knowledge was fixed during its training, months before your conversation. Your exchanges do not change it live. If you have not turned the option off, they may be used to train a future version, mixed in with millions of others: this is a question of confidentiality, not personal memory.

Key points

• Context: the current conversation, emptied when a new one is opened.
• Persistent memory: notes fed back in by the software, viewable and deletable.
• Training: fixed before you; your conversations do not change it live.