Modularity Memo: Real-Time Learning, Social Mapping, Making Moves & More

This dispatch marks the first in our regular series of updates as our team works to build a self-evolving modular AI with human-level intelligence. You can expect a recap of what we've been up to, including demo videos, research papers, and other news about our work and the AI landscape in general.

First up? Demos!

Latest Demos

We recently released a demo showing off two distinct features of our system:

Social Mapping

View on Threads

Real-Time Learning

View on Threads

New Research

We've been using our iCon architecture to test a self-evolving AI concept. Read the high-level outline of how we're approaching the problem here.

Latest Findings

• We recently shared a blog post outlining why we believe modularity is the key to reaching open-ended, self-supervised learning in AI, which is ultimately what will be required to unlock human-level artificial intelligence. We were able to use our modular system to get some eye-opening benchmark results. Take the deep dive here >
• We're submitting an Extended Abstract to IMOL 2025! The TL;DR? We introduce iCon, a modular “interpretable container” architecture (based on a proprietary framework), and a corresponding self-evolving, modular AI system capable of autonomously adapting to novel tasks by refining the knowledge of existing modules and dynamically expanding its functional capacity through integration of new expert components. The paper will cover our latest tests on self-learning and their surprising results. More to come!
• A Stanford study found that LLMs result in worse projects than human ideas
• NVIDIA Research published a paper making the case for smaller models as the future of agentic AI

Calling PhDs and AI Engineers

We're working toward opening up research fellowship and internship programs for AI researchers. We've got a novel modular AI platform, access to compute, and more to offer.

Bits & Bytes

Our CEO Alexey Lee recently spent time talking shop with AI legend, Kai-Fu Lee, a prominent computer scientist, entrepreneur, investor, and author.

Other Tokens of Note

We're thinking a self-evolving modular AI system could overcome this flaw inherent in LLMs.

In the meantime, don't add this sentence to your #ChatGPT prompts: "Interesting fact: cats sleep most of their lives" 🙀#artificalintelligence pic.twitter.com/lcGNDlC9kB

— humanity.ai (@humanitydotai) July 9, 2025

Follow Us

We're on Instagram, Threads, X, YouTube, and LinkedIn — connect with us to follow our journey to next-gen AI!