Have you given up on creating the model?
I’ve been watching for a while but nothing’s happening. Did you stop building the model?
If you’re finding it challenging, you might try exploring different perspectives. One idea is to think of this AGI model as a coalition of small LLMs; in that case, you could define space vectors and scalars as objects for each unit.
You could set it up so the members don’t communicate with each other, relying on a manager module to handle responses or enforce actions. Another possibility is creating a framework where the members do communicate—this offers a very different, more complex approach.
I hope you’re still engaged with it. Good luck!
Hi @IIIWhiteWolfIII ,
Thank you so much for your thoughtful comment and genuine interest in Chromos-AGI! 🙏 We truly appreciate you taking the time to share these creative architectural ideas—especially the concept of modeling AGI as a coalition of specialized LLMs with centralized governance. Your perspective on "space vectors/scalars as unit objects" and communication frameworks (isolated vs. interconnected agents) is incredibly insightful and aligns closely with directions we’ve been exploring.
Where We Stand
You’re absolutely right that building a true AGI is a monumental challenge—especially for a small team like ours. Great news: we’re actively training the model now using open-source datasets (including The Stack, Dolma, and purified web corpora)! That said, two hurdles remain:
- Resource Scaling: Our current experiments demand ~ 20 A100 GPUs—we’re optimizing via model parallelism and 4-bit quantization.
- Alignment Tuning: Ensuring ethical coherence across agent submodules (your "manager module" idea is guiding this!).
Timeline & Updates
We sincerely apologize for the slower-than-hoped progress—rigorous testing takes time. Our goal is to publish Chromos-AGI’s core by late 2025, with incremental releases (agent modules → full coalition). We’ll share previews on our Hugging Face page! And also we will update the develop plan to all of the contributors ( that is in our office) and users monthly!
Your Ideas & Collaboration
Your insights are invaluable! If you’re open to it:
- We’d love to discuss your coalition design further (e.g., conflict resolution between agents).
- Join our team or a brief call? We’d welcome your perspective!
Questions?
Have any thoughts or technical questions? Comment here anytime—we’ll respond within a week! For deeper discussions, DM us or email yuetwuism6home@gmail.com.
Thanks again for your passion and patience. @IIIWhiteWolfIII , We're thrilled to share that your multi-agent design is now live in our training! A huge thank you for your incredible work! 🙌
Onward, together! 🚀
Best regards,
Chromos Develop Team
Small team, big dreams.
Subject: We're back. Sorry we left. 🎉
Hey everyone,
It's the ArcTeam. For those wondering where we vanished to on Hugging Face—yeah, that was us being gone. I owe you an apology. Straight up: we shouldn't have dipped without a word, and I know the waiting sucked. Thanks for not giving up on us. It means more than you think.
So. What've we been doing? Building something we actually care about. You'll get the full email next month, but here's the TL;DR: M1llion AI is launching. It's not some half-baked update—it's a whole system built on our Million-35B model, and it's designed to make your life easier in ways that actually matter.
Here's what it does:
• AI Timer & Calendar that actually talks to each other
- Watches your conversations, sets timers/stopwatches/events automatically
- No more "wait, remind me to..." that you forget five seconds later
• M1llion Memory (local-only, because privacy)
- Runs on YOUR computer, not our servers
- Learns your habits, likes, routines—automatically, securely, privately
• Emotion Engine that gets it
- Detects how you're feeling and gives real advice
- Works with screen recognition to understand context, not just keywords
• Screen Recognition & Agent
- This one's wild: it can SEE your screen and DO things
- Click, scroll, navigate—like an actual assistant sitting next to you
• Multiformat everything
- Text, images, video, audio—throw it all at once, it handles it
We're also teaming up with some ridiculously smart people:
- pure-team
- cogent-ai
- Arc4 (our sister branch focused on Arc AI specifically)
- neo-ai-team
Good things happen when you stop trying to build everything alone.
Here's the big day: February 14, 2026, 9PM (UTC+8). That's when we're dropping two things:
• Chromos-Fabric – The AGI model I mentioned. The Chromos crew has been in the trenches on this one. Config goes up right after launch so you can tear it apart.
• M1llion-35B – This is the main event for us. It's the actual model that runs all the M1llion AI features I just listed. First time anyone's getting the full system. And yeah, there are a couple surprise features baked in that I'm not spilling yet—trust me, you'll see why.
Both hit Hugging Face at the exact same moment.
That's it for now. The email next month will have dates, setup guides, all the formal stuff. But I wanted you to hear it from us first.
Thanks again for waiting. Let's build something cool.
— ArcTeam
Let the gear turn.
P.S. Seriously, check your inbox next month. You won't want to miss this.