[bsenftner]

Although my work is not about loneliness, it has similarities with your goals. My work is about creating intellectual confidence, critical awareness, and laying the foundation for honest ambition from self confidence through accomplishment.

I've been creating AI chatbots, and "taskbots" (chatbots that do more than respond conversationally, they procedurally do things on request.) These are all embedded into an office software suite such that it forms an office software environment with dozens of virtual expert co-workers that are integrated right inside the UI of the office software. So when one is working, there are multiple virtual co-workers that are conversationally inside the same software you are using, with access to what you are doing in that software. They advise your work, they can directly manipulate one's in-editor work, and in general they are designed to educate you how to do your own work better, how to understand past your work and become materially better at what one does.

As I created and have been using and testing the system with general office workers, I find I need to include psychological aspects in the AI Agents behaviors, because people are intimidated, or they are sarcastic, or they are really timid and afraid of doing something wrong and getting reprimanded. The AI Agents that help a person edit in the work processor require instruction that people are both afraid to reveal any lack of understanding, and have a real hard time articulating the help they need. So these word processor support AIs coax the user and coach them how to ask for help; once that hurdle is crossed, users get active and chatty with the agents and make good progress. That initial use, they are very intimidated. Plus often confused, because they think they can just say "write this for me" and the AI will do everything, as if it can read their mind.

If you find this interesting, you can contact me at https://midombot.com/b1/home

[omars_]

Are your users engaging with your various AIs through text?

I didn't see any place I could contact you from the page that you linked.

When you say people are intimidated, sarcastic, timid, or afraid, are you measuring that or just observing it personally as users try out the app?

The techniques you're applying around coaxing / encouraging certain behavior could apply more broadly, depending on how you're managing it.

[bsenftner]

I have both text inputs and voice, where the voice converts to text before submitting to the LLM. That allows people to edit their voice transcription, and so on.

Oh, you'd have to create an account. That's free, and nothing happens with the email you give to create the account beyond use for password recovery.

The behavior of users, their reactions to the site, are both from observations and my asking them, and them telling me. I'm writing it at a law office where it's in use by the attorneys. Turnover in staff gives me a pretty good idea how a fresh set of eyes looks at it, and as I've improved things that feedback is fresh from new people.

I'm making it as broad as I can at the moment, seeking to find a balance between automation and interactivity that promotes creative flow. I'm trying to do with writing literature, spreadsheets and generalized project management what people are doing with code AI integration. This includes students learning how to use these applications, as well as advanced users of them, but not necessarily programmer types, nor people comfortable with the idea of getting that technical.

A lot of what I find I need is communications establishment between a user and the AIs they have access. They don't know what to say, how to ask for information, how to basically use them. Then, for example, when they do ask the AI something that ask is loaded with implied context that AI does not know nor could know. The ask the AI to do their work, without explaining what that work is, or what the expect from the AI in response.

So I have added interfaces for specific and direct use, such as here is a location where you can ask the AI to do editing changes to your document, and over here is a place where you can ask the AI about the quality of the writing. Each of those are specific knowledge bots pre-loaded into that part of the interface, one can edit the document in all kinds of ways, and the other can give feedback on it from all kinds of perspectives, like a writing professor or coach, so one writing it can comprehensively write better.

Each of these AI integrations with one of these tools is also conversationally programmable. So a user can have a series of them, each with different knowledge, for different scenarios. Those then get collected into similar AI groups I'm calling organizations, because they end up working in tandem with each other and the user.