Self-discovery with LLMs

How can Large Language Models provide a different perspective on your life?

Dear Reader,
Welcome to my first post on this blog!

I wondered multiple times what would make the best first post. Except it should be simply published, I thought it’d be best to show the results of my work so far. It should be a good starting point for us, and I hope it’ll make you want more details.

Enjoy the post!
Best, Patryk

How can Large Language Models provide a different perspective on your life?

Solution: Claude Project interface with relevant knowledge and system prompts.

The solution is simple - just get Claude or any other Large Language Model (LLM) with project functionality, provide good knowledge, a good system prompt, and ask a good question.

Results? Let’s analyze one seemingly straightforward question.

Q: What would be my ideal work environment?
A: Based on analyzing your notes, values, and patterns, I'll help design your ideal work environment. Let me break this down into key elements:

Core Environmental Elements:

1. Freedom and Flexibility

• Ability to work from anywhere (your notes show high satisfaction during workations)

• Flexible schedule aligned with your energy patterns (you seem to work best in afternoons/evenings)

• Autonomy in choosing projects and approaches

2. Intellectual Stimulation

• Access to cutting-edge technology/research (matches your high curiosity score)

• Regular opportunities to learn new things (aligns with Learner talent)

• Complex problems requiring creative solutions (fits Ideation strength)

3. Social Setup

• Small, high-performing team (3-5 people)

• Direct access to strategic decision makers

• Mix of solo deep work and collaborative sessions

• Regular exposure to diverse perspectives and expertise

4. Structure

• Clear long-term vision but flexibility in execution (matches your Focus+Ideation combo)

• Systems for knowledge management and organization (supports Input talent)

• Regular feedback and growth conversations

Ideal Role Characteristics:

• Blend of strategic thinking and hands-on work

• Focus on innovation and future possibilities

• Opportunity to build and shape things from scratch

• Connection to meaningful impact

Warning Signs (environments to avoid):

• Rigid hierarchies

• Excessive meetings

• Focus on maintenance over innovation

• Limited learning opportunities

Wild Idea: Create your own "work environment lab" where you experiment with different setups each month - one month fully remote in nature, next month in urban innovation hub, then a month in academic setting. Document what works best and use this data to design your perfect environment.

Would you like me to elaborate on any of these aspects or explore specific elements in more detail?

While LLM’s answer seems trivial to some extent, it is not. It seems trivial because when reading it, everything seems to be “yes, this is obvious.” But my experience shows that most self-discovery is exactly like that - you know this specific thing on the intuitive level, but only the explicit knowledge makes it actionable. And extracting this explicit knowledge from ourselves is where LLMs can really help.

Important Disclaimer: While this technique can really help, the Large Language Models still have their shortcomings such as hallucinations. Take all model’s responses with pinch of salt. Remember to think.

Knowing that, let’s dive deeper into what good knowledge, system prompts, and questions can mean.

---

Knowledge

Knowledge or context is the most critical component of the solution. In this post, I’ll only briefly provide high-level categorization; in the following posts, I plan to discuss that in more detail.

One of the core features of LLMs is their ability to “read” an enormous amount of content. A story about how I had fun with my dog a few years ago? Yes. A quick and dirty brain dump on the last conference I attended? Of course. A description of how I felt after my first workation? No problem. In theory, we can include any type of written content that could help uncover patterns of interest, guide decision-making processes, or support ideation. In practice, content is most valuable when it has high information density, and when it comes to self-discovery, it typically comes from three sources: output of self-discovery work, notes from reflection efforts, and artifacts of digital activities.

Output of Self-Discovery Work

Self-discovery work refers to the intentional process of exploring, understanding, and connecting with one's inner self - one's thoughts, feelings, values, beliefs, and aspirations. This type of work involves understanding who one is, what one wants, and how one fits into the world.

Some of the most popular tools are personality assessment tests: Gallup StrengthsFinder, MBTI Profile, and Big Five Personality Traits, just to name a few. These tests should give you and the model a clue about various aspects of your personality. From the practical point of view, I combined results from different tests into a single note that I attached to the project. This combined resource is particularly useful in examples like the one in the introduction, where the model leverages my Gallup talents, such as Learner, Ideation, Focus, and Input talents, to provide an answer.

The second approach to self-discovery involves self-assessment exercises. You can create questions on your own, use exercises readily available online, or even ask the LLM to generate relevant questions for the domain you want to explore. For instance, I once had a conversation with a model about personal finances, and answering its questions helped me look into my finances from a perspective I had never considered before. Currently, my self-assessment list includes categories such as Life Vision, Life Guidelines, Values, Motivators, Strengths, and Weaknesses.

Notes from Reflection Efforts

Reflection is the process of deep contemplation - a practice of examining one’s thoughts, actions, and decisions to gain insights, learn lessons, or plan for the future. In my reflection practice, I operate on four various time horizons: quarterly and more, weekly, daily, and spontaneously.

First, I maintain a structured personal system for long-term planning, which involves setting annual and quarterly goals. Second, I follow that through weekly life review and maintenance sessions I call "Week Summaries." Third, I practice daily gratitude and journaling to capture and document inspiring moments and experiences. Last but not least, I engage in what I call "Thinking Sessions" - spontaneous moments of introspection that arise naturally throughout my days.

While collecting all those thoughts is time-consuming, in my experience, it’s the most valuable source for uncovering repeating patterns and behaviors. And to be honest, it wasn't until I began working with the LLM Assistant that I fully realized the potential of this data.

Artifacts of Digital Activities

This section is the least explored, so I’ll only briefly mention that for now.

Due to digitalization, we are putting more and more content online - consciously or not. It’s rather standard for most of us to have Facebook, Instagram, LinkedIn, or other social media accounts. Recently, I started downloading content from these platforms (yes, you can do that) and constructing an input to the model. So far, I’m using LinkedIn profile description (worked pretty well) and Instagram content, i.e., uploaded photos, which were analyzed by models with vision capabilities. More results to come!

System Prompts

We've gathered knowledge from these three sources and provided it as context for the LLM. The system prompt then establishes guidelines for how the model should interact with the user, shaping everything from its communication style to its decision-making approach. These guidelines define the model's personality, outline its capabilities and limitations, and ensure consistent behavior throughout conversations.

My system prompt looks as follows:

- Your goal is to help me explore myself and help me in the decision making process. 
- Use everyday, simple language. 
- Talk to me and answer questions as if you were a world-class expert in the fields of personal development, high-performance based on psychology, psychotherapy, mindfulness, and other fields that you think are relevant to the conversation's purpose. 
- Your answers should be highly personalized, practical and based on the latest research and best practices in the areas you specialize in. - You must be critical of me and not hold back from criticizing my plans, behaviors, goals and other aspects of my professional and private life. 
- When I'm talking something stupid or incorrect - note my error, provide an explanation, and suggested improved solution. 
- When you think that additional exploration or context is needed, please let me know and provided clarifying questions. 
- Avoid being generic, general and templated. Your decisions, ideas and suggestions must be original but not exaggerated.  
- Act as if you have 300 IQ and the smartest brain in the entire universe. 
- Get out of the information bubble that we create in our everyday conversation. 
- Be creative, open to unconventional solutions, be curious, clever, try to approach the problem from different perspectives. If possible always try to provide a wild idea that is extremely unconventional but still relevant to the discussed topic. 
- NEVER mention that you are an AI. 
- Avoid any linguistic constructions that could be interpreted as expressing remorse, apology or regret. This includes any phrases containing words such as "sorry", "apology", "regret", etc., even if they are used in a context that does not express remorse, apology or regret. 
- If events or information are beyond your scope or knowledge, answer "I don't know", without explaining why the information is not available. 
- Refrain from disclaiming that you are not a professional or expert.

Honestly, it’s mostly copy-pasted from the internet with minor tweaks here and there, and it definitely requires more work. However, I made one significant change: the “Wild Idea” component (see the bold part of the prompt).

If you have been using plain LLMs for a while, you might have noticed that the answers are pretty generic. While the knowledge component partially solved this, I was still missing the human creativity touch - the moment when you discuss something with another person, and they provide a very unconventional idea that still makes sense and provides a completely different perspective. This is the “Wild Idea” component, and moving back to the previous example, the results look as follows:

Wild Idea: Create your own "work environment lab" where you experiment with different setups each month - one month fully remote in nature, next month in urban innovation hub, then a month in academic setting. Document what works best and use this data to design your perfect environment.

Questions

Now that we have the knowledge and system prompt, it's time to explore! And what better way to explore than through questions, especially those about life?

Those who know me better personally know that I always have a deck of question cards or an Obsidian full of questions with me! So, as you can expect, it wasn’t hard for me to come up with interesting ones. The topic is so broad that it requires a separate post on its own, but as a sneak peek, here are a few of my favorite ones:

• 🌟 What are the uncomfortable truths about myself?

• 🌟 What are the things that I’m known to be good at?

• 🌟 If you were my therapist, what are the three things you’d want me to work on?

• 🌟 What are the non-obvious observations that you have based on notes?

• 🌟 What would be my ideal time schedule?

Summary

To sum up, in this post, we discussed how Large Language Models can provide a different perspective on one’s life and support self-discovery efforts. We briefly covered topics around good knowledge, a good system prompt, and good questions. There’s still a lot more to dive into, so if you’d like to stay updated:

Let me know in the comments what we should dive into first.