I Hadn’t Mentioned My Dogs

I asked ChatGPT about owls. The answer came back with a message about my dogs. I hadn’t mentioned my dogs. IT can’t connect those dots. AI can. That is a glimpse into AI memory and the power of inference.
By Jon Nordmark, CEO, Iterate.ai

What you'll learn

•How AI links things you said in separate conversations, without being asked
•Why that linking (e.g. memory) is what makes AI useful — and what to govern
•What you can take back once a model has learned it, and what you cannot

The core argument

In June 2025, thinking about the owls perched above my neighbor’s deck, I asked ChatGPT what kind of owls live in my neighborhood. It said probably a Great Horned Owl. Then it brought up things I hadn't asked for. It mentioned Rosie, Sammie, and Charlie — my dogs.

The exchange, and two of my dogs — The actual exchange — and two of my dogs taking a rest.

“

AI will soon have the “infinite, perfect memory” no human has — and today we are only in “the GPT-2 era of memory.”

Paraphrasing Sam Altman, CEO, OpenAI — Big Technology Podcast, Dec 18, 2025

Now think about those names. One in seven people use a pet’s or family member’s name in a password (UK NCSC; Yubico, 2025). Street and kids’ names answer most security questions. Your AI collects all of it while being helpful. A password has to be stolen. Memory only has to be asked.

That is inference: the model combined things it already knew about me and predicted a future action. Everything useful about AI starts there — and so does everything you have to govern.

Where that picture lives

Your AI creates a profile of you. Where it lives is a decision, in a public, shared, or private environment.

NetApp | Iterate.ai

I Hadn't Mentioned My Dogs · 01 / 02

How it knew

Weeks earlier, in a different conversation, I had mentioned my dogs by name. The model kept that. When owls came up it put the two together on its own: owls hunt small animals, small dogs go out in the evening, this person has small dogs. AI built that connection.

Traditional IT cannot do this. A search returns what matches your words. It carries nothing from last month into today’s question. That difference is the point of AI — and the risk.

Two Great Horned Owls on a roofline — Two Great Horned Owls, roosting on a neighbor’s roofline.

What this means at work

The upside

An AI that links institutional data knows more about how your company works than any one employee does.

The catch

It builds those links from what your people typed. A question from Finance can surface a document from Legal without anyone deciding it should.

The part that sticks

You can delete a file. You cannot un-prompt what a model already learned.

NetApp® AIPod Mini keeps your memory and inference private, behind your firewall

Inference is not the problem. Where it happens is. The AIPod Mini runs the model, so the semantic relationships it builds stay on your hardware and under your own access rules.

Two words worth pinning down

Inference: working it out

— reaching a conclusion nobody told you. You see a leash by the door and know there is a dog, even though no one said so. AI does this constantly, across everything it has been told.

Memory: what it keeps

— what the system retains after the conversation ends. Some fades when you close the window. Some is written into the model itself, where you cannot reach in and take it back.