Building an AI Coach to Tame My Monkey Mind
eugeneyan.com
eugeneyan.com
I would think twice to give my soul to an unregulated tech company on the quest to monetize, and I seriously advise anyone to do the same.
It is amazing how often people let themselves be screwed over in exchange for convenience or quick results.
More broadly, I feel I cannot use AI unless it runs locally and nothing would ever leave my device.
Edit: I also value the freedom to be able to decide what works for me and what doesn't.
Just kidding, of course. I agree 100%. More problematic, I think, is that everyone is using it. So, somebody else might put some sensitive information about you into the AI. Kind of similar to people posting pictures on social media without your permission.
Or uploading their full contacts to any chat app.
They call it Human Nature.
Especially one whose CEO is known for a cryptocurrency scam.
https://www.technologyreview.com/2022/04/06/1048981/worldcoi...
https://www.buzzfeednews.com/article/richardnieva/worldcoin-...
AI therapists have no such patient confidentiality regulations, so it would be way easier to do that.
Imagine if OpenAI wants to get rid of a problem employee who happens to use a OpenAI-powered therapist.
1. I don't need more technology in my life and more detachment from human connection. If I need a therapist, I'll work with a human.
2. I don't trust companies with such sensitive and personal data.
RE 2nd point, this comment is spot on - https://news.ycombinator.com/item?id=39999142.
I would never use an external system for this. Even ignoring the privacy issues, and they are plentiful, it makes a lot more sense that you tailor the AI therapist to your needs, which most of the time are not generic needs. If you're in a therapy program you probably know those needs better than most people and can concretely ask the AI to assist you in meeting them over a longer period of time.
I'm happy with my system, it's a little clunky still but it is doing the job in assisting my recovery, which is all I want.
I built an AI journalling assistant plugin for Obsidian that has the option for 100% local and private use via Ollama and the model of your choice.
https://github.com/clairefro/obsidian-chat-cbt-plugin
Not a replacement for human interaction, but finding it actually really helpful for getting out of negative thinking spirals in between therapy sessions. Plus it has endless patience :P
I feel LLMs work well as objective sounding boards. Trying to build something similar to the OP for this reason!
I think other way to view it is, AI ecosystem is growing so fast that almost anybody can have a good idea and glue some thing together and be the first.
Guess it's Tomato Potato.
At one point in time just putting data in a grid was a pretty simple idea, and thus we have Excel dominating the world.
I'm not sure we can say that any 'obvious' idea right now wont turn out to be huge.
Seems like there is a tendency right now to crap on anything to do with AI, even if it is a nice little app, that really isn't a bad idea even if it doesn't deal with privacy issues.
Some experiential inquiry exposing the extent to which vast swathes of "suffering" are purely the result of overidentifying with a meaningless stream of thoughts would go a lot farther than building an AI coach that can only perpetuate engagement with same.
> My secretary watched me work on this program over a long period of time. One day she asked to be permitted to talk with the system. Of course, she knew she was talking to a machine. Yet, after I watched her type in a few sentences she turned to me and said: ‘Would you mind leaving the room, please?’
https://theconversation.com/my-search-for-the-mysterious-mis...
It reminds me alot of the AI therapist from the „pod generation“ movie. Frankly the voice got better and building sentences, but the underlying personalised part is just not there.
My approach is more how can I make a todo list app that’s not pushing hustle but also not ignorable.
https://github.com/CominiLearning/recenter
Often find myself mindlessly clicking over to reddit or some other site while thinking and before I know it I am being sucked in by something there. Those minutes add up. Wanted something that proactively tells me that this is happening and also summarizes how time was spent. Have found it be a huge help!
Some of the more standard LLMs try very very hard not to insult me and tell me to talk to a licensed professional for everything. That's not what I want. I wanted to get the most impolite LLMs I could, so I looked up some list about which ones were the best for topics like erotic roleplay and those worked way better for even non-sex topics because they're not RLHF'd to be unconditionally nice or excessively cautious.
It sounds like a bad relationship and you should break up with them.
Putting in "Why am I not a billionaire yet?" is a lofty enough goal that I am not beating myself up too badly for not meeting that goal. "Why am I not a world class athlete for my age group?" could be another one.
Knowing and trusting people who care about you. Building a support system. Asking them. Going to a support group (online or not)
Not by detaching yourself from reality and asking self help questions to a statistical number generator probably made by some of the largest tech companies around. (With a consistent track record of causing and profiting off mental health problems and addictive dark UX patterns)
If you add up costs of the entire stack: Vapi + Deepgram + GPT4 + ElevenLabs, it comes to be $18/hr
That's not exactly cheap.
That's 52 weekly one hour "therapy" sessions for the price of one month of weekly sessions with a human.
Effectively, what you are looking at is three things:
1. Voice transcription.
2. An LLM using the transcription as input with a custom system prompt.
3. Text to speech to return the text output as speech.
For the first point you could look into OpenAI's whisper. Whisper is open source so you can run it locally in theory, although I don't quite know what the requirements and performance is there.
The second point can be anything, really. If we keep it to being open, Ollama with a capable enough model can fit the bill.
The third TTS bit I don't quite know to be honest. I have no clue what the quality is of open TTS speech engines, but I am pretty sure there are a bunch out there.
Running everything locally probably does require a pretty beefy GPU to be able to run it all in a competent manner.
Once you have access to all three, making something like a coach should be fairly easy. Though, fairly does a lot of heavy lifting there. It entirely depends on your coding capabilities and front-end knowledge.
One thing you need to decide on is how portable this needs to be. Mostly because of security and authentication if you want to access all of this from a phone outside your local network. Assuming you want to be able to access it everywhere, I'd maybe go for a single endpoint API that accepts audio files as input and also gives audio as a response. Then on the backend you can access the various components.
You need to create a UI for voice that facilitates a voice call. Having a continuously open microphone is a bit tricky because then you need to figure out stuff like detecting pauses to send the audio off. So you could for a press to talk method to simplify things.
You need to figure out the prompt. Also, how much history you want to keep. This depends on the context window the LLM you are using is capable of.
If you just want the LLM to respond, then this is actually easy enough. The conversation you can then simply store for reading it back later. You then need to pass the output to the TTS engine you are using and send that back to your front-end for the response.
Doing all of this with the OpenAI API is also possible and would simplify a lot of things. Although, I'd probably still go for my own API and handle accessing the various OpenAI endpoints in the backend.
Personally, I have been thinking about it lately. Though without the TTS aspect back, as I am not looking for a coach but have been brainstorming about an implementation that looks at my ramblings and then makes notes, reminders in my calendar and other automated actions. Basically something like your regular Google assistent/Siri/Alexa but without the vendor lock in and tailored at my own tools.
So far I have kept it at brainstorming, though. But it does mean I did look into the tech stack you might need, so maybe it is helpful to you :)
As soon as it has released the last (concept) branch, it is grasping the next... Like a monkey swings through the trees.
> It is a Buddhist concept that describes a state of restlessness, capriciousness, and lack of control in one's thoughts.
In the quiet jungle of the mind, Where thoughts intertwine and bind, Lives a creature, restless, undefined, "Monkey monkey," whispering, unconfined.
"Monkey monkey" in the trees, Swinging wild, in the breeze, Grasping branches of worries with ease, Never still, never at peace.
This monkey leaps from limb to limb, From dawn till dusk, on a whim, Thoughts to thoughts, the world so dim, "Monkey monkey," the mind's hymn.
In the canopy of consciousness it dwells, Through the forest of thoughts, it propels, "Monkey monkey," it spells, A tale of unrest it tells.
But hush, now listen, there's a way, To calm the monkey, to sway, With mindfulness, day by day, "Monkey monkey," will no longer stray.
Breathe in deep, let it go, Watch the monkey, but don't follow, In the mind's eye, let it show, "Monkey monkey," now moves slow.
In the quiet jungle of the mind, "Monkey monkey," has resigned, Peace and stillness, intertwined, The restless creature, now aligned.
"Monkey monkey," once did roam, In the heart, it found a home, Within the calm, no longer alone, "Monkey monkey," in tranquility's dome.
Tiger got to hunt, bird got to fly;
Man got to sit and wonder 'why, why, why?'
Tiger got to sleep, bird got to land;
Man got to tell himself he understand.