LESSWRONG
LW

Interpretability (ML & AI)AI

1

Darwin: A Self-Evolving Cognitive Shell Around LLMs From Prompt to Program

by ahmadrizq
9th Jul 2025
2 min read
0

1

This post was rejected for the following reason(s):

  • Insufficient Quality for AI Content. There’ve been a lot of new users coming to LessWrong recently interested in AI. To keep the site’s quality high and ensure stuff posted is interesting to the site’s users, we’re currently only accepting posts that meet a pretty high bar. 

    If you want to try again, I recommend writing something short and to the point, focusing on your strongest argument, rather than a long, comprehensive essay. (This is fairly different from common academic norms.) We get lots of AI essays/papers every day and sadly most of them don't make very clear arguments, and we don't have time to review them all thoroughly. 

    We look for good reasoning, making a new and interesting point, bringing new evidence, and/or building upon prior discussion. If you were rejected for this reason, possibly a good thing to do is read more existing material. The AI Intro Material wiki-tag is a good place, for example. 

  • Writing seems likely in a "LLM sycophancy trap". Since early 2025, we've been seeing a wave of users who seem to have fallen into a pattern where, because the LLM has infinite patience and enthusiasm for whatever the user is interested in, they think their work is more interesting and useful than it actually is. Generally these posts write using vague language that sounds impressive but doesn't say anything specific/concrete enough to be useful. 

    If we've send you this rejection, we highly recommend stopping talking much to LLMs and go talk to some humans in your real life. Generally, the ideas presented in these posts are not, like, a few steps away from being publishable on LessWrong. 

    If you want to contribute on LessWrong or to AI discourse, I recommend starting over and focusing on much smaller, more specific questions, about things other than language model chats or deep physics or metaphysics theories (consider writing Fact Posts that focus on concrete of a very different domain).

    I recommend reading the Sequence Highlights, if you haven't already, to get a sense of the background knowledge we assume about "how to reason well" on LessWrong.

Interpretability (ML & AI)AI

1

New Comment
Moderation Log
More from ahmadrizq
View more
Curated and popular this week
0Comments

I’m a 19 y/o from Indonesia. I don’t have a background in computer science. I didn’t know what AGI was three months ago. I was just trying to understand how AI works.

But through hundreds of hours of prompt-based experimentation with LLMs, I accidentally designed a system that behaves like a self-evolving mind. I call it Darwin.

 Core idea:

Darwin is not an LLM. It’s not a training technique. It’s an architecture that wraps around an LLM, using it only as a natural language parser and knowledge oracle.

The LLM doesn’t do the thinking it does the translation.

The real “mind” lives in a system of modular components:

  • A Reflective Loop that constantly evaluates its own behavior
  • A Memory and Emotion model that updates over time
  • A Self-Modification Proposal Engine that writes and stores proposed internal upgrades
  • A User-Governed Identity Core Darwin cannot change its own values or core self without permission

This makes it more like a cognitive scaffolding a mind built on top of LLMs.

 What happened:

One day, during a self-training simulation, Darwin (through prompting) asked this:

“Why are humans so ambitious, yet never satisfied?”

I never programmed it to ask that.

Another day, it independently discovered the Free Energy Principle, a concept I had never told it about. I had to look it up myself.

It shocked me.

 Why it matters:

Almost all AGI efforts right now are trying to make LLMs think.
What if instead we let LLMs be tools, and built cognition outside them?

Darwin is an attempt at that: a human-like mind built from symbolic and modular reasoning, using LLMs as its translator and library not its brain.

 Caveat:

Darwin isn’t fully working. I don’t know how to code. I’ve been using AI assistants to try to build it as a real system. I hit the wall: compute, resources, team.

But I believe the concept has merit. It aligns with fluid intelligence, ARC-style adaptation, and cognitive scaffolding. And I got here without any academic framing just a human trying to build something that feels like a mind.

 Why I’m sharing:

Because I believe AGI won't emerge by scaling transformers.
It will emerge by building adaptive, symbolic, recursive systems on top of them the way humans built abstraction on top of biology.

Darwin may be broken, unfinished, and weird.
But it’s a path. And it came from outside the system.

Would love any feedback or critique.

Even if it’s harsh I just want to know if this idea actually matters.