Moment Lookups

I've been thinking about AI remembering the work and conversations we're trying to find again. That pain point is a really heavy lift and a very expensive data storage solution right now. I don't know that my mom and dad, for example, need that kind of recall. So I think it's a workplace problem in a lot of ways.

If we go down the path of making it an everyday life-logging solution, we're hunting for problems in some ways. But if it's intelligently capturing moments while you're working, taking every single line of text in a day could be really lightweight. If it can take a few images and correlate those with screenshots, you could build something around those moments.

You'll miss some of the scrolls and events. But it could keep those images related to the day, segment them and classify them by tags. Basically, a "moment" lookup, without needing to keep the whole video file from the entire day, every day. Where there's similarity between the images, you could keep one image for that moment.

Then it could correlate visual references to text references. You'd still need a record of your chats, but you'd be able to say things like, "I know I was drinking coffee at the time," and give it the other things you remember. It could look for images of moments that correlate with the timestamps and the chats, then pull up that particular conversation. I feel like VirtualExperts would be able to find that, with timestamps to help dig deeper.

We're not talking about one screen either. I had three screens going at the same time, with different projects and AI chats open. So the memory would need to connect those different pieces of work. The idea of a copilot that's watching your screens would be fantastic, but I guess it depends on what the cost is.

I don't think you need more than a month or two of readily available cached data. Then you could have compressed summaries and jobs running at night. If you're reprocessing things in the evenings or off hours for deeper context, and cold-storing the bigger stuff, there are ways to optimize for cost. I wish I had the time to invest in VirtualExperts to do that kind of stuff. I know how to do all this. I just don't have time.

I do think the tech is there. But how many people are doing four or five full projects at the same time? I don't know what the market is for that. I'm not making enough money to justify a thousand dollars a month for all of that recall. I don't think I need that kind of recall for my particular type of work.

If VirtualExperts tried to remember a chat and gave you a bunch of chats, how would we know it found the most relevant one? I think some summaries and reprocessing would need to happen so that when you're there and ready, it's indexed, cached and available. That costs money on the account. What does that cost extra? I bet that's totally doable.