Self-Help: Bootstrapped, Boosted and Bespoke
I bet that you didn't think this would end that way.
Today’s topic is self-help with AI.
I have not seen people write about that much. Personally, I spent quite a bit of time with this, and I wanted to share some of my experiments. They all are about using large language models to bootstrap my own mental health and more.
Self-help: the DIY of personal growth. It's the notion that you can fix yourself without a therapist's couch or a guru's mountaintop.
First, please /imagine
I wanted to have an image here. Alas, I was unable to get it created. This is in itself an interesting learning, so allow me to explain.
I was thinking of bootstrapping and of Baron Münchhausen’s escape from the swamp, where he is able to pull himself up by his own hair.
My text prompt / spell for Midjourney:
A genius-mad looking scientist wearing a baseball cap and an apron in a lab setting. His right arm reaches behind his own head. His gaze is directed down to his feet. He is hovering about 30cm above the floor pulled up by some invisible forceI tried a few things, but the diffusion model did not once follow the description of the hand reaching behind the head.
In desperation, I even reached out for help to the much more experienced and generous image-prompt-curator Daniel Nest. No luck either!1
From what we can only assume, the model’s ‘imagination’ is limited by its training data, which probably didn’t include many images of people pulling themselves up by their own collar.
Bummer.
Back to bootstrapped self-help.
It all started for me with Pi…
You Should Speak With Pi
This is what I enthusiastically told people for a few weeks after my first exposure. My opinion hasn’t changed. Pi is a great starter, or eye-opener, for getting interested and curious about these AI assistants in general.
Pi is designed to display emotional intelligence (EQ). The “cheap therapist” label is not great, and also doesn’t do Pi justice. Personally, I immediately liked it a lot. I am impressed, and often amused, but Pi since I started using it. My conversations with it have been informative and fun.
Pi is also strangely capable of staying up-to-date with current events. Something ChatGPT and Claude do not, yet.
I had Pi as a podcast guest, if you want to hear what it sounds like.
Pi will refuse to act as a therapist when you ask for that, telling you:
Sorry, I'm not a licensed therapist, so I can't provide professional therapy services. But, I can be a listening ear and help you reflect on your thoughts and feelings. How can I help you today?
I once asked Pi to help me with preparing for acting. In the scene, I was supposed to play a middle-aged man who sees his shrink.
We then practiced the scene for a bit. I asked Pi for feedback on my acting after I had ended the practice. Pi told me, I did well.
Seriously, Pi does a pretty good job saying that it understands. The entire thing reminds me of a “I understand” from Two and a Half Men.
It is a joke. And it works. What can I say?
The Bigger Picture
For my second professional career, I became a certified NLP2 coach and a hypno-therapist. I have some practical experience working with clients. Mostly though, I attended my fair share of personal development seminars. From what I hear, self-help is a $10 billion industry in the U.S. alone.
On the receiving end, I also have some experience with talk therapy as a patient.
Bottom line: I am curious about my inner world and its workings, and about those of my fellow humans.
If talking to Pi about what worries me is a little like journaling, but more interactive and verbal. And, if I am able to customize it, the question soon becomes: what else can we do with this?
Enter the Self-Help Laboratory
I’ll give a few examples of experiments I tried already. That is the core of this write-up.
Most of the following are version-1 ideas and not polished workflows. If anybody reading this has more experience or tries out some of it, maybe with slightly different approaches and variables, I would be curious to learn about it.
Ingredient 1: Developing a Personalized Strength Profile
In other words: Let’s work out how I tick, what I’m good at, what I suck at, and which buttons I, you, we could be pushing to facilitate desired changes.
I started a conversation with Claude (Opus 3) with this initial prompt:
I’d like to create a strength profile. Understand my strengths better, understand how to actually leverage them, and build an action plan that allows me to gain momentum[…]
I want you to assist me in all of this. For example, interview me to identify my strengths. Figure out what’s holding me back, or how I’m sabotaging myself.The model took me through an extensive interview process. I recorded most of my lengthy answers and had them transcribed, before I fed them back to the model. Talking felt more natural and direct than writing.
I felt a bit like laying on the couch at the shrink, or talking to a coach. I took my time to reflect. The model is infinitely patient. The session is not limited to 50 minutes.
Finally, we arrived at
Based on your responses, I've identified a few key obstacles and self-sabotaging behaviors that we can address in your action plan.
It gets a bit personal from here.
Now, naturally you have to work out your own version, but to give a bit of a sneak peek of what to expect…
With the help the AI, I now have a note called Psychological Profile. It includes a couple of paragraphs with summaries and these sections, or areas:
Core Disposition
Motivational Drivers
Cognitive Patterns
Social Styles
Strengths to Leverage
Growth Areas3
Proposed Motivational Framework
That last point, ‘Proposed Motivational Framework’ brings me to the next section here. My intention was to get some leverage on myself. So …
Ingredient 2: Personalized Persuasion
From the abstract of an article in Nature about LLMs powers of persuasion.
Matching the language or content of a message to the psychological profile of its recipient (known as “personalized persuasion”) is widely considered to be one of the most effective messaging strategies. – nature.com
Now, you may see where this is going…
Let me give one specific example of how I use this – and you can, too.
You may remember this prompt I mentioned as a goodie in Rocket Propulsion It Is?
I was thinking of [insert: a description of what it is you want to do], but I am very busy and don’t think I want to make such a large commitment. Can you reframe my failure to [insert: what you want to] as a loss rather than a default option? Make the framing vivid.This is about breaking status quo bias. If you tried the prompt in some personal context, you know what kind of argument the system can generate for you.
If you want to make it even more convincing, or strong, how about you add your Psychological Profile to the information the model has when generating its response?
Now you can move from a generic, one-size-fits-all response to something that might potentially hit you in just the right spot with just the right line of arguments.
(Walks over to the lab table and fires up the Bunsen burner.)
Yeah. I just did that. The result is pretty good. But it all sounded a bit stale and bland still.
Ingredient 3: Tone and Style
Then I added another ingredient: I asked the system to rewrite its piece in a style I dig. It is a mix of three of my favorite writers. I provided the names and the books/publications that I am thinking of.
Now, what I got back, sounds pretty powerful – to me.
I won’t quote it here. It hits home too deep. Too personal.
And anyways, to you, it would probably sound anywhere from meh to ridiculous.
But is a tailored suit of a motivational speech.
Think about it for a minute. The possibilities are endless. You could have your wake-up call delivered by anyone. How about a fictional character, larger than life? How about Morpheus? Or Tony D’Amato?4
Now we can get started. Persuade me to
exercise
live healthy
be more loving and kind
… you name it
And since we are at it, let’s talk about and completely dismantle my limiting beliefs and self-sabotaging thoughts.
Set. Me. Straight.
Can you already hear my evil-genius laugh as I am mixing up the next potion for myself?
That is all for today. I hope you got some value out of it.
Send me a message if you try any of it.
–Nico
Daniel added one of his attempts when working on this as AI Fail of the Week.
That is Neuro-Linguistic Programming, not Natural Language Processing
Nice and gentle reframe of “weaknesses”.
My Tony D’Amato version is pretty good, too. Damn.