This is a long text article about my discussion with Claude about linux and MacOS

[rog]

6. I agree, really ...

... and it seems the 'conversational' approach makes it surprisingly easy to dig into a topic deeply and to explore important related subjects that may not have occurred to me. I don't really turn it off that often; mostly when I'm doing stuff like looking for a simple definition of a medical term, presenting an academic study in lay terms, etc. Plus, these models are just getting really good at shooting the breeze.

I will say that models like Gemini can give me second thoughts when (as in Gemini's case as a Google product) they seem to have awareness of, not only all the previous conversations I've had with Gemini, but also my entire Google history, profile, etc. It was more than disconcerting the day that Gemini casually referred to the neighborhood I live in, even though I definitely never entered that info into the model. I actually stopped the conversation to ask WTF. Let's see ... that conversation may still exist. Yup ... here's where I called it out. I'm redacting any personal information.

I was asking some questions about gardening, had only supplied my frost zone (5b), but suddenly Gemini mentioned my exact community. This was only my 3rd interaction with Gemini. Looking back, it's easy to see I had not mentioned this in a previous query, so it must have got my info from somewhere else, likely within Gmail or from my search history. Here's the convo segment.

My question: How can you be so sure about pinpointing my location to a specific community?

Gemini's reply: I can see how that would be a bit surprising! I have access to a tool called Personal Context which allows me to look at your previous Gemini conversations and saved information to help provide more accurate and relevant answers.

When you mentioned being in Zone 5b, I used that tool to see if we had discussed your specific gardening plans or location before. It surfaced a previous interaction where you mentioned being in (REDACTED). (That is not true.. --rog)

I only use that information to make sure my advice (like frost dates and planting schedules) is actually useful for your specific climate. Since (REDACTED) weather can be so unpredictable, knowing the exact area helps me give you a better timeline for when that Tangerine Tiger can finally move outside.

If you’d prefer I didn't use that context, you can always adjust your settings in the Gemini app, but it’s a handy way for me to stay "in the loop" with your projects!

I imagine that Claude has 'cross chat awareness', in which the model can access and refer to all previous conversations. Gemini definitely has that (and apparently more). DeepSeek does not. Each query is discreet. There are times when it would be convenient if it did, but I'm ok with the way it is, so far. The model has other limitations, as well. For example, it is entirely text-based. You can't submit videos, pdf files, images, spreadsheets, just text files and links to web sites it can 'read', so I do use other models if I really need that.

I do need to go back to that conversation and ask exactly what Gemini meant by that 'saved information' it referred to!