Memory
nounWhat an AI product saves from earlier work and adds to later work. Many products keep it for the person using them, not for each client.
On Monday, a mortgage adviser asks the firm's AI product for an email to Leanne Price about pausing her house purchase. Leanne has just been made redundant, so the adviser asks for a gentle tone. On Wednesday, the adviser asks for an email telling Mark Ellis, another client, that his remortgage application has gone to the lender. The draft says the firm knows this is a difficult time after his redundancy. It goes out unnoticed. Mark, who has been in the same job for ten years, rings to ask what redundancy the firm means.
The AI model, the part of the product that writes, keeps nothing between pieces of work. The product around it can. On Monday the product saved a line in its own words: the client has been made redundant, so keep the tone gentle. It kept the line for the adviser, not in Leanne's file, and added it to the context of everything the adviser asked for afterwards. In Mark's email, the AI took "the client" to mean Mark. That saved line is memory. Memory is what an AI product saves from one piece of work and adds to the context of later work, leaving the model unchanged. The product chooses what to save, sometimes without being asked.
A firm already keeps notes for later work: the file note. It sits in one client's file and goes into work on that client only. The comparison breaks on where the note is kept. Many products keep memory for the person using them. Unless a product keeps a separate memory for each client, one client's details can turn up in work on another.
Not to be confused with
- The context is the material the AI has for one piece of work. A memory affects the work only when the product adds it to the context.
- Within one conversation, the product sends the conversation so far to the model with each new message. That is part of the context, not memory. A new conversation starts without it, apart from anything the product has saved to memory.
- Training is how the company that built the model made it, and it is not staff training or CPD. Saving a memory leaves the model unchanged. Whether that company also trains later models on the firm's conversations is a separate question, answered in the product's settings or the vendor's terms.
- Model, in this entry, means the part of an AI product that writes, never a model portfolio.
Overheard
Why did Mark's email mention a redundancy? It was Leanne who lost her job.
The product saved a note about Leanne to your memory, not to her file. Delete it in your settings. I'm switching memory off for the firm until the vendor can keep it per client.
Where it sits
Copy this entry
Paste it into Word, Google Docs, Notion or Apple Notes. Headings and links come with it.
Sources
Your feedback
Could you explain memory to a colleague now?