"Why SymbolicAI" is the Wrong Question

[ software , neuro-symbolic ai ]

“Facts do not cease to exist because they are ignored.”
—Huxley

Look, I get it. You want to know why SymbolicAI is the future, why it’s the next big thing, why it’s going to change everything. You want to know why it’s better (or worse) than LangChain, why we don’t have X or Y, why we don’t just add X to make it better than Y, why we don’t make more video tutorials, or why we don’t make it more accessible.

But here’s the thing: those questions miss the point I care about, and are ones I don’t entertain. I would argue that those “whys” are a category error. It’s like asking why the sky is blue when you just want to know whether to grab an umbrella before going out.

I want to share something George Hotz said in a recent stream that is exactly what I mean—and what prompted me to write this post:

ANON: Maybe put why tinygrad makes sense over other libraries?

GH: I don't like doing things like that, and not to go off into the weeds into like philosophy, but I'm not trying to tell you why tinygrad is right for you. That's what advertising is. That's a form of manipulation, and it doesn't go anywhere. It's a race to nowhere. My point is, we don't do that. I'm not trying to sell you on using tinygrad; I'm trying to tell you what tinygrad is. The best I can do is tell you what it is, so you understand it, and then you fundamentally have to make your own decision about whether you want to use it or not. You know, I explain this philosophy to people, and people think I will lose because of it, and if I do lose because of it, well so be it, because the minute you start doing the other thing you already lost. You lost the minute you started playing. People in general don't make good decisions; they like being sold something shiny. [... proceeds to say the parable of the Chinese farmer …] It comes down to time horizons. People might like being sold something shiny over a short-term horizon, but if the world continues trying to just sell shiny things to people, the people who fall for the shiny things will end up wireheaded. If you don't retain an ability to independently evaluate, it will be a complete ratcheting up until very powerful forces have entirely manipulated you.

What I’ve always done is show you how cool SymbolicAI is, how it can be used, and how it can be applied to solve real-world problems. I’ve met people that I—to a fair extent—convinced to at least try SymbolicAI, and they all ended up doing fine in their day to day development. If you’re a developer, I invite you to reserve a “discovery” weekend for this framework, where you go down rabbit holes with childlike innocence. If it’s not for you, that’s fine. I’m not here to convince you that SymbolicAI is better than anything else. I just showed you what it can do and let you decide for yourself. This isn’t to say I don’t want you to use SymbolicAI. I do. But I want you to choose it because you see its power and potential, not because I’ve perfected a sales pitch.

Now, there are tools like DeepWiki that provide better documentation than I could ever write, and I don’t want to compete with that; I’ll correct it where it’s plain wrong. Please go read SymbolicAI’s DeepWiki breakdown. There’s a lot of interesting stuff in there. For instance—and to my knowledge—this framework is the first to bring Design by Contract principles to the world of LLMs, which I think is a big deal. All the applications I shared after March were built with contracts—usually chained together, at most 3 to 4 at a time—and I think this is the way forward because it just works.

I’m working on new ontology research. Once open-sourced—I expect the next month, this project will hopefully be a great introduction to contracts. Everything from ontology creation to knowledge graph generation and natural language to Neo4j Cypher translation is one modularized pipeline held together by contracts. The deep research agent I built maxed out at 580 sources in one session in about 40 minutes for roughly $1. Here’s a sample output, where I gave it this query: “my wife has tinnitus and i want a comprehensive exploration of tinnitus relief techniques”. I find existing deep research tools—from OpenAI, Google, and Perplexity—extremely lacking in comparison. I’m not saying it’s better, or bulletproof, but different in a way that I think is more useful. All these were enabled by SymbolicAI’s contracts.

The sad reality is that very few people have written to me, to say nothing of how many have contributed. Most of the messages I get are from people who are already using it and liked it, which aren’t that many either. All the silence makes me feel like I’m in the Dark Forest, giving away free ice cream while preaching sermons, ready to be neutralized at any moment.

Anyway, my DMs are open—feel free to tell me I’m completely off base.