Suggest edit — Empowerment Algorithm App

Title

Name

Note

---
confidence: high
related:
- wiki/concepts/wu-wei-disclosure.md
- wiki/concepts/engagement-vs-empowerment-algorithms.md
sources:
- raw/transcript.md
title: Empowerment Algorithm App
type: project
visibility: public
---

# Empowerment Algorithm App

Jacob's "secret open-source project" for shifting the [[Wu Wei Disclosure|dopamine-to-calorie ratio]] of his own attention.

## What it does

> "I made a secret open-source project that filters my X and Google News feed and other feeds, conceivably too. And it filters it per a very fast local LLM, so it removes the low-vibration stuff — only high vibes."

Two layers:

1. **Local-LLM feed filter** — runs over X, Google News, other feeds. Removes "low vibration" content based on learned signal. Keeps only "high vibes."
2. **Live productive/unproductive labeling** — keystroke shortcut while browsing tags any site as unproductive. Metadata is pulled and used to train an underlying model on productive vs. unproductive distinction.

Over time, the model generalizes well enough that Jacob doesn't need to label anymore. **And then unproductive websites auto-close.**

## The exact mechanics

> "I just put a thing today where I would just go on with my regular day browsing the internet, and then just by keyboard shortcut, I can say, okay, this website is unproductive. And then it will categorize that website as unproductive and pull all of its metadata to train an underlying model on what unproductive looks like, what productive looks like."

> "And then you can pretty much just do this for like any website. And at a certain point, you don't even need to train ahead anymore — it just is so intuitive that it just predicts which websites are productive and unproductive for you."

> "And the key is, like, for unproductive websites, it just instantly closes them. As soon as you open an unproductive website, it just closes the tab."

The closing-behavior is the load-bearing part. **Filtering content** without removing the consumption affordance still leaves the dopamine path open. **Closing the tab** removes the path.

## Why "wu wei"

This is [[Wu Wei Disclosure]] in software form. The app doesn't require willpower; it changes the **gradient**. After it's been trained, the cheap-dopamine path is closed off; the path of least resistance now points toward something productive.

## Related work

- **Harrison's "pause" app** — Jacob mentions Harrison built a sibling project, framed as a break app rather than a closer. Same family.
- **AI as competing dopamine source** — Jacob's adjacent observation: AI (specifically vibe-coding with Claude / Cursor / etc.) provides an *alternative* high-dopamine path that happens to be productive. So the empowerment app removes the bad path *and* the AI provides a good path; both shifts compound.

## Why "secret"

Jacob says "secret open-source." The transcript doesn't elaborate. Plausible reasons:
- Personal-use first; not yet documented for distribution
- Some secret-sauce prompts / models he doesn't want immediately scraped
- Just work-in-progress

David expressed interest in collaborating: "Dude, I would love to collaborate with you on something like this."

## Why this is high-leverage

The reasoning chain:
1. Most "willpower" failures are gradient failures ([[Wu Wei Disclosure]])
2. Modern feeds are designed to make the wrong gradient default
3. A small intervention that changes the gradient at the OS level dominates a thousand "be more disciplined" interventions
4. If population-scaled, this is a plausible intervention against the [[Engagement vs Empowerment Algorithms|engagement-algorithm pathology]] David named as the single biggest current problem ("80, 90% of high-schoolers and children are addicted to social media")

## Related

- [[Wu Wei Disclosure]] — the underlying mechanic
- [[Engagement vs Empowerment Algorithms]] — the comparison
- [[Harrison]] — built sibling "pause" app
- [[Sparks of Motivation]] — what cleared attention surfaces