I am looking for good options for an elderly relative which has some onset of slight dementia, but is still living alone (with assistance).

The main problem is that usual things are becoming just too complicated for that person. There are too many details that don’t matter. For example, user interfaces have a lot of features that are not strictly needed, many hidden states, often no clearly visible affordances, and so on. Some mail clients make it difficult to recognize phishing.

As a good example how it is done better, I’d name GNOME simple-scan. A really good and simple UI!

This person is currently using Thunderbird + Firefox on Windows, and K9 Mail on Android. Learned to use Windows in their sixties and is now in mid-eighties age.

The main need is a simpler Email client.

Any suggestions?

  • SavvyWolf@pawb.socialEnglish
    2·
    17 hours ago

    Google, Microsoft, banks and everyone in between has been spending the last decade rolling out sophisticated scam detection systems and infrastructure. If a scam can get through all that, then it can get through your little kiddy model.

    And even if you did have a model that worked well, why is the best interface for it an llm? Why not just have a big red “scam warning” banner? Loud, clear and to the point.

    AI agents can make mistakes, that’s something you’re going to have to explain to the user. But you also have to explain to the user to always listen to the scam warning even if their gut thinks otherwise. How do you define whether or not you should trust the AI in such a way that someone with cognitive impairment can apply it?

    • sudoer777@lemmy.mlEnglish
      3·
      15 hours ago

      If a scam can get through all that, then it can get through your little kiddy model.

      I use FastMail which has a reputation for decent spam filtering, and spam messages still fall through. I also have AI agents process my emails and it correctly detects those messages as spam.

      why is the best interface for it an llm?

      Because you can talk to it like a person and it’s not bounded by a GUI’s features. Compared to humans they’re more readily available and you’re not risking your relationship if you need to ask something you wouldn’t want to ask a person

      Why not just have a big red “scam warning” banner?

      Because we already have those and they aren’t reliable enough

      How do you define whether or not you should trust the AI in such a way that someone with cognitive impairment can apply it?

      When in doubt ask a human when one becomes available