Not for journaling or therapy, but almost every time I draft something in German, I ask a model to "rate my German" so that I keep improving.
Before chat models, I did something similar with Google Translate, but GT doesn't tell me why things should be different; and worse, GT tends to auto-correct typos, leading to stuff like this: https://translate.google.com/?sl=de&tl=en&text=Schweinwerfer...
("Schweinwerfer", spelled like that, would actually mean "pig thrower": Schein (shine) vs Schwein (pig))
Not exactly on purpose, but I used Grok for some business planning, and now it sometimes loops that into other answers, suggesting how something else I asked about could overlap with the business project, or just asking at the end, "So, would you like to work on [other project] some more?" It's a little reminder that I wanted to get back to that project, which doesn't hurt since I tend to procrastinate or get distracted and forget about projects.
I think to be a real "coach" they'd have to be able to push you proactively. You can script that kind of thing, but if you're getting that done, maybe you don't need much pushing.
The proactive piece would be key for me as well. My challenge right now is to find the right setup for leveraging my journals and notes and a local llm for this (don't want to share such personal data with openai, google, etc.)
A friend is using ChatGPT as a kink and poly aware pseudo therapist, keeping in mind the limitations of the system, considering the challenge of finding a human version tuned for such a lifestyle.
Not for journaling or therapy, but almost every time I draft something in German, I ask a model to "rate my German" so that I keep improving.
Before chat models, I did something similar with Google Translate, but GT doesn't tell me why things should be different; and worse, GT tends to auto-correct typos, leading to stuff like this: https://translate.google.com/?sl=de&tl=en&text=Schweinwerfer...
("Schweinwerfer", spelled like that, would actually mean "pig thrower": Schein (shine) vs Schwein (pig))
Not exactly on purpose, but I used Grok for some business planning, and now it sometimes loops that into other answers, suggesting how something else I asked about could overlap with the business project, or just asking at the end, "So, would you like to work on [other project] some more?" It's a little reminder that I wanted to get back to that project, which doesn't hurt since I tend to procrastinate or get distracted and forget about projects.
I think to be a real "coach" they'd have to be able to push you proactively. You can script that kind of thing, but if you're getting that done, maybe you don't need much pushing.
The proactive piece would be key for me as well. My challenge right now is to find the right setup for leveraging my journals and notes and a local llm for this (don't want to share such personal data with openai, google, etc.)
A friend is using ChatGPT as a kink and poly aware pseudo therapist, keeping in mind the limitations of the system, considering the challenge of finding a human version tuned for such a lifestyle.
Interesting, no worries with privacy when using chatgpt?
This is not a concern they have. YTMMV (your threat model may vary, n=1). I offered to wire up ollama or deepseek for them and they shrugged. If you want to host local, consider https://github.com/OpenHealthForAll/open-health and https://github.com/johnbean393/Sidekick as patterns. Please consider open sourcing whatever you build.