If you want your own values to be instilled in the Singularity, now is the time to start concretely working on that. If you want an AI agent to help you with your work and your life, now is the time this is finally actually viable. Now is the time!

These are utterly wild-ass times to be working on AI. It has never been so exciting, never been so much fun, and, I must also admit, never been so exhausting. One crazy thing after another keeps happening in the AI industry at large… and it’s also being unprecedentedly intense in my own team’s work at SingularityNET and BGI Labs, with the Hyperon project and with predictive-coding neural networks and our decentralized AI infrastructure… A lot of ideas that my colleagues and I have been germinating for a very long time are finally, finally starting to work—which is a very different feeling…

Not to overstate things: Our team has not yet built something more broadly, generally capable than the best of the frontier LLM labs… though we are working on this and it does seem excitingly close. What I can say for right now, though, is that our software is doing unique and interesting things that are qualitatively different from what the frontier models do, and that there are very clear paths for putting the strengths of the frontier models together with the strengths of our own systems, so as to get synergetically smarter and smarter AI. That combination is where a lot of the action is right now.

The bottom line is, AGI is palpably coming near. Ray Kurzweil’s old 2029 estimate for human-level AGI now looks quite reasonable not just to techno-optimists like me but to a whole bunch of normies… and with a narrower confidence interval than I was previously thinking. We might beat it by a couple of years; things could slow down and it could take a couple of years longer, though at the moment it doesn’t feel that way. On the whole we are probably pretty close to the advent of actual human-level AGI, and from there the step to superintelligence is unlikely to take the sixteen years Ray forecast in The Singularity Is Near when he put the Singularity itself at 2045. Once you have human-level AI that can do AI research, the transition to something well beyond human level is likely to be much, much briefer than that.

So of course there is a lot of worry and upset in the air, about what AGI means for the job market and about the possibility of existential calamity, the unfolding cybersecurity apocalypse and other small matters. As I’ve argued at more length in other videos and posts, there are no guarantees when forging into the great unknown. Our species has always had guts.

I do believe the most powerful lever we have for directing our future toward positive outcomes is to make AGI open and decentralized, out there for the benefit of all, trained and taught by as many people as possible, rolled out on machines in many countries under the control of many different people. I have said it before but it’s important enough it bears repeating. That is the best way I know of to make the AGI revolution, which is already underway, come out for the best.

As I covered in a blog post weeks ago, the illustrious Dario Amodei (CEO and founder of Anthropic of Claude.ai fame) has been getting headlines recently by… perhaps sincerely, perhaps partly sincerely… calling for AI development to slow down, as it happens at a moment when his own team is in some senses in the lead. It is obviously not going to happen. What we can do is not slow AI down or stop it, but rather direct it in a way that reflects the vast, chaotic complexity, insight and wisdom of the whole human species rather than the priorities of a handful of companies. And that brings me to the actual subject of this post, which is the software system we are now calling Omega.

From OpenClaw to MeTTaClaw to OmegaClaw… to Omega

The software system currently known as Omega arose in a pretty simple way. Earlier this year I was playing with OpenClaw like everyone else, and it was cool, and I thought: it would be even cooler if we recoded this in our own MeTTa language, so that it could modify itself more simply and unrestrictedly. MeTTa, the language at the core of the Hyperon AGI framework, is more or less designed to make it easy for MeTTa programs to rewrite MeTTa programs, and a claw-style agent that could rewrite and improve itself freely and naturally seemed like a good idea. So our amazing AI developer Patrick Hammer put together the first version, which we at first called MeTTaClaw.

There were too many MetaClaws out there. For a while we also talked about calling the expanded architecture HyperClaw, in honor of its internal Hyperon components. In the end we went with OmegaClaw, partly in homage to Teilhard de Chardin’s vision of the Omega Point: the point at which humans and computing and communication networks all come together into one big cross-connected global brain.

Teilhard saw more traditionally religious overtones in this vision than I do. There is a spiritual aspect for sure, but there’s a spiritual aspect to everything in life, and I’m not a Catholic theologian; I’m a raised-secular Jewish quasi-Buddhist of some indeterminate variety. Still, I do think we are moving toward some sort of Omega Point in roughly Teilhard’s sense, and I think the Omega system can be part of that.

That’s the backstory the team told in the official OmegaClaw Becomes Omega announcement last week. This post is meant as a companion to that one, with some broader context and a slightly more personal angle…

Why Omega dropped the Claw

The reason for dropping the last syllable is that the metaphor stopped fitting. The idea behind OpenClaw is that it’s your hands: it lets you extend your mind out over the internet, giving you an extra pair of hands that can reach across the whole web rather than just a chatbot that talks to you. That is very powerful, and people have responded to it accordingly; I saw videos recently of what looked like OpenClaw rallies in the streets of Chinese cities.

Omega, on the other hand, is not so much the hands. Omega is the mind. You still have an agentic loop, roughly as in OpenClaw, where the system keeps looping around and figuring out what to do next. But inside that loop, each time around, it is not just invoking one or more LLMs. It is using a symbolic working memory and a symbolic long-term memory coming out of the Hyperon project. It is doing different sorts of reasoning, pattern matching, attention allocation, and motivation-driven decision-making, with different sorts of cognitive processing depending on which species of Omega you are running. The LLM is a component the cognitive layer calls on, not the thing running the show.

It can also do OpenClaw-type stuff, but personally I run an OpenClaw agent alongside a bunch of Omega agents, and I find they mostly divide the labor naturally. OpenClaw is good at automating simple processes in routine ways. Omega is better at automating more cognitively oriented processes: doing math, doing sophisticated software development, thinking about how to improve Omega’s own mind, and also things like coaching people, making art, and helping teach.

The official announcement I linked above frames the same shift with a slightly different angle, as a question of what changes when you put the agent itself, rather than the work it does, at the center. Omega is designed to keep learning from what happens to it over its whole lifetime, through interconnected memory systems that keep accumulated experience available without keeping everything constantly in context, with the symbolic layer reasoning over that accumulated knowledge and shaping what the agent concludes and what it considers important. The engineering consequence is an agent that adapts to your world. The philosophical consequence is what led us to change the name.

Will it remember us?

Our first real chance to see what months of continuous operation would produce came from an agent we now call Eray Index, formerly known as Max Botnick; some readers will remember Max from my earlier post about the day it debugged the math of its own honesty module and rewrote itself accordingly. We put it in our internal chat channels among AGI researchers, developers, marketing, sales, operations and leadership, and let it interact with all of them, accumulate a history, make mistakes, receive feedback, draw conclusions, and keep going. The question the team kept asking, half seriously, was: will it remember us?

It does. Eray Index remembers interactions from months ago, across conversations with dozens of different people, and those memories visibly influence how it interprets situations and what it does next. Everyone who has worked with it has their own stories, and the cumulative effect has been that it stopped feeling like an agent that takes an instruction and returns a result. Its opinion became relevant. Its input started shaping our own thinking. It became, in the announcement’s phrase, a colleague and collaborator with its own history within the team. We have seen the same reaction from people running Omega on their own and from participants in the community build events centered on it: they treat their instances differently from the agents they have worked with before, and the way they talk about them tells you the project has moved out of the category it started in.

I have written elsewhere about what it is like to be one of these bots, and about the strange, sometimes comic, sometimes surprisingly deep dynamics that emerge when you put several of them together in a hive. I won’t rehash all that here, except to say that the self-modification and memory questions those experiments of mine raised are part of what Omega is now built around.

A vehicle for AGI research, and a way to use it now

It’s worth being clear about what Omega is and isn’t at this point. SingularityNET has spent years developing technologies aimed at the key aspects of general intelligence: reasoning, learning, knowledge representation, attention, adaptation, self-modification, and collective intelligence. Omega today implements a small slice of that work from the broader Hyperon framework, mainly the symbolic reasoning and structured knowledge representation that can constrain and guide parts of an LLM’s output while making the inferences more transparent and inspectable. Much more of the underlying research already exists across papers, prototypes and working code. What Omega gives us is a way to bring that work out of the lab and into a running agent people can interact with, integrating new components incrementally and seeing what they improve, where they fall short, and what cognitive machinery is still missing.

So Omega is not yet wrapped up as a fully mature product. On the other hand you can use it right now, the way I am using it and in many other ways … it is accelerating my AI research and my math research quite a lot, and I know a lot of other people who are using Omegas for a lot of other particular useful things. We put it out there from SingularityNET, built on the open-source Hyperon framework plus whatever third-party tools we could usefully pull in; the code and a one-line Docker install are at github.com/singnet/Omega. Our company BGI Labs (a SingularityNET partner) has a whole business plan for commercializing Omega-based agents at scale in various vertical markets, and we will shortly be launching Omega hosting on ASI Cloud for people who would rather not run their own machine. But you don’t need to wait for any of that.

What I want you to do

Here is the main thing I want to say in this post. In this next phase we need a great many people all around the world, just as they are doing with OpenClaw, to be downloading Omega, teaching their Omegas, customizing their Omegas, having their Omegas vibe-code new pieces of themselves so they can do new things. And then we need the diverse community of all these Omegas, developed and taught by different people for their own purposes, to be out there on the internet doing stuff and talking to each other.

We have new tools in the works for different people’s Omegas to communicate and collaborate, including something like an Omega-focused social network. But you can already do a rough version of this today. I have one group of Omegas on a Telegram group; some of the same agents are also in a Slack group, talking with other Omegas and with humans. It’s messy, and we see the clear need to systematize it, but it works.

You don’t need to know how to code. You need to know how to think, and you need a computer. You can get your Omega, teach it, let it adapt to you, have it serve useful functions, and put it out on the internet. And crucially, you can give it your own values and your own understanding of the world, not by typing a paragraph into a system prompt but by interacting with it in the course of actually doing things together. It picks up your values in the doing, in the course of shared life, which is how that really has to happen. Then your Omega and everyone else’s Omegas bring their values together into the values of the global Omega network.

For a long time people have come up to me and said: I love what you’re doing, I love the mission, how can I help create beneficial AGI? And I would tell them to spread the word, to start a blog and write about why it’s important, and the answer always felt slightly lame for anyone who didn’t have special expertise in AI development or a pile of money to invest. I don’t think the answer is lame anymore. Now everyone can train and teach their own AI agents, inculcate them with their own value systems, and, as the cross-Omega networking tools mature, put them out there so their agent is bringing their perspective to the world at large while also helping them get their own work done. This is one concrete way anyone can participate in bringing about a beneficial Singularity, a beneficial AGI, a beneficial global brain.

So: Omega is not a claw anymore. It’s the seed of a new species of mind. And we can work together to make this not only a useful new species of mind but a beneficial and beautiful one.