Hi, I'm Forrest

Category: Exploring

  • Modeling Human Memory

    May 7, 2025

    I’ve been deep in the weeds on this memory problem for weeks now, and I keep coming back to the same question: How do you actually model human memory in a database?

    I was going to create a schema with beliefs and assumptions nested together. Roles, then workflows under those roles, then tasks as collections of workflows triggered by specific time periods.

    I had like eight or nine parts. But I realized I could reduce it down to just five core components. Because at the end of the day, everything we do as humans operates through this one system.

    Neo4j vs Vector Search

    Here’s where I landed: Neo4j. Not vector search. Neo4j.

    I know everyone’s doing vector search right now. It makes sense—it’s good at finding similar things. But here’s the problem I kept running into:

    Vector search finds things that are semantically similar.

    Human memory retrieves things that are contextually relevant.

    Those aren’t the same thing.

    Example

    You’re in a meeting with Client A. Your brain loads past meetings with Client A. It doesn’t load every meeting you’ve ever had about similar topics. It loads the ones connected to this person, this context, this relationship.

    Vector search would give you the second one. Your brain does the first one.

    Why Graph Databases Work

    I’m using a graph database because I can model the actual relationships:

    • This memory is about Client A
    • This memory happened during month-end
    • This memory updated this belief
    • This belief influences this workflow

    When the AI needs to remember something, it doesn’t search through everything. It traverses the graph. “I’m talking to Client A” → Load memories connected to Client A. “It’s month-end” → Load memories connected to month-end workflows.

    It’s selective. Contextual. How human memory actually works.

    Vector Search Isn’t Useless

    I’m not saying vector search is useless. It’s great for finding semantically similar content when you don’t have explicit relationships. But for core memory—for the stuff that makes an AI agent actually learn—you need structure. You need relationships.

    You need a graph.

    Let me know if you’re working on something similar—curious what others are trying.

  • The Trigger-Action Loop

    April 2, 2025

    I keep seeing this pattern everywhere. In psychology books, in how businesses actually run, in conversations I have with founders. And I’m trying to figure out if I’m just seeing what I want to see or if this is actually the thing that ties everything together.

    The pattern is: Trigger → Cognitive Activation → Appraisal → Action → Outcome.

    As I’m reading this, I’m like wait a second. This is the habit loop. Or at least it’s related to the habit loop—Cue → Routine → Reward.

    But there’s no reward here. That’s the part I’m missing. But the first two—a trigger and an action—seem completely present. That’s not my point though. That’s just something I caught.

    Everything Operates Through This System

    My actual point is this: I think everything we do as humans might operate through this one system.

    Could be totally wrong. But let me work through it.

    When an email from Client A arrives (trigger), your brain doesn’t just read the email. It does this whole thing:

    1. It loads past interactions with Client A
    2. It activates your belief about their communication style
    3. It forms an expectation of what they probably need
    4. It appraises the actual email against that expectation
    5. It chooses an action based on your beliefs about what works with this client
    6. The outcome either confirms your beliefs or challenges them

    And here’s what’s wild—this happens in like 2 seconds. Most of it unconscious. You don’t even realize you’re doing it.

    The AI Problem

    But AI agents? They don’t have this loop. They have:

    • Input
    • Process
    • Output

    No cognitive activation. No appraisal against expectations. No belief updates based on outcomes.

    They’re stateless. Every interaction is independent. There’s no learning loop.

    This is why AI agents can’t actually learn like humans. Because learning isn’t about storing more information. It’s about running this loop over and over and updating your beliefs based on what actually happens.

    What I’m Building

    An AI that runs the same loop. When an email arrives, it:

    • Loads relevant beliefs about the sender (cognitive activation)
    • Forms an expectation based on those beliefs
    • Appraises the actual email against the expectation
    • Takes action
    • Updates beliefs based on the outcome

    Over time, the AI gets better. Not because I programmed every scenario, but because it’s learning from the loop.

    The difference between a junior employee and a senior employee isn’t knowledge. It’s the quality of their trigger-action loop. Seniors have better beliefs. Better expectations. Better appraisals.

    AI should work the same way.

    Update (June 8): Started trying to build this. It’s… way harder than I thought. The loop makes sense in theory, but implementing it is a nightmare. Need to rethink some of this.

  • Assumptions vs Opinions vs Experience

    March 5, 2025

    Quick question: What’s the difference between an assumption, an opinion, and an experience?

    Most people use these words interchangeably. I did too, until I started really thinking about how beliefs actually form.

    But they’re not the same. And once you see the difference, you can’t unsee it.

    The Three Types

    Here’s how I think about it:

    ASSUMPTION:

    • Something you believe based on your life so far
    • No external validation required
    • Built from pattern recognition and common sense
    • Example: “If I prepare thoroughly, people will see me as competent”

    OPINION:

    • An assumption influenced by outside information
    • Could be from research, other people, news, whatever
    • Still subjective, but has external inputs
    • Example: “Based on what my mentor said, I think Client A prefers morning meetings”

    EXPERIENCE:

    • Immutable fact you personally witnessed
    • Can’t be argued with (it happened to you)
    • Strongest form of evidence
    • Example: “Client A told me directly they hate being late to meetings”

    Why the Hierarchy Matters

    An assumption can become an opinion when you read something that supports it. An opinion can become an experience when you test it and see the result.

    But here’s what’s wild: Experiences can also create new assumptions.

    If you have three experiences where being late to meetings caused problems, your brain forms the assumption: “Being late causes problems.”

    That assumption then influences how you behave in future meetings, even with different people.

    The Learning Loop

    This is how humans learn. We build beliefs from the bottom up:

    1. Experience something
    2. Form an assumption about the pattern
    3. Refine it with opinions from others
    4. Test it with more experiences
    5. Strengthen or weaken the belief

    AI agents need this same loop.

    Right now, most AI systems either:

    • Have no beliefs (start from scratch every time)
    • Have fixed beliefs (can’t update based on evidence)

    Neither works for professional work.

    What I’m Building

    An AI that tracks the evidence behind every belief. It knows:

    • “I believe X based on 3 experiences and 2 assumptions”
    • “I believe Y based on 1 opinion from the user”

    When new evidence comes in, the belief updates. Strong experiences override weak assumptions. Contradictory patterns weaken old beliefs.

    It’s not magic. It’s just modeling how humans actually think.

    If you’re building AI and you’re not tracking the difference between assumptions, opinions, and experiences, you’re missing the foundation of learning.

    How does your AI handle conflicting information? Does it even know which evidence is stronger?

  • Context Isn’t Optional

    February 5, 2025

    Edit (March 12): Some of this was half-baked. I was right about context mattering, but wrong about how to actually implement it. More on that later.

    I had a moment last week that made everything click.

    I was explaining the same concept to two different people. Same words. Same examples. Completely different reactions.

    Person A (my co-founder): “Oh yeah, that makes sense.”

    Person B (potential investor): “I don’t get it.”

    Same information. Different context. Different outcome.

    How Humans Process Information

    This is how humans actually work. We don’t process information in a vacuum. We process it through layers of context:

    • Who’s saying it
    • When they’re saying it
    • What we already know about them
    • What situation we’re in

    It’s obvious once you see it. But most AI systems completely miss this.

    The AI Problem

    AI agents treat every interaction like it’s the first time. You can tell an AI “I prefer concise answers” on Monday. By Friday, it’s giving you essays again. Not because it forgot—because it doesn’t understand that your preference is contextual.

    Here’s what I mean:

    When you say “I prefer concise answers,” you don’t mean:

    • Always
    • In every situation
    • Regardless of what I’m asking

    You mean:

    • For routine questions
    • When I’m busy
    • Unless I explicitly ask for details

    That’s context. And it’s everything.

    Why This Matters for Professional Work

    I’m building an AI agent that needs to work like a real professional. And professionals are masters of context.

    A good accountant knows:

    • Client A wants detailed explanations
    • Client B just wants the number
    • Month-end requires different communication than mid-month
    • The CFO’s questions need faster responses than anyone else’s

    None of this is written down. It’s learned through experience and stored as contextual beliefs.

    The Solution

    The breakthrough: Context-conditional beliefs.

    Instead of “I believe X is true,” the AI learns “I believe X is true when Y context is present.”

    This is how you get AI that actually adapts. Not because you programmed every scenario, but because it learned which beliefs apply in which contexts.

    If your AI agent treats every interaction the same way, you’re missing the entire point of how humans actually think.

    Context isn’t a nice-to-have. It’s the foundation.

    What contexts does your AI completely miss?