The hardest problems in artificial general intelligence, and our attempts to solve them.
Stay Updated
Get a monthly update when we publish new findings, problem breakdowns, or solution attempts.
Updates
We are opening up our research log. Follow along as we tackle the hardest problems on the road to AGI.
Talos has been an internal research project at Atheneum AI for over a year. We have learned a lot, hit many walls, and found some interesting ways around them.
Starting today, we will share monthly updates covering the problems we are working on, the approaches we are trying, and what we have learned along the way.
About
Talos is our active attempt at building artificial general intelligence. We are developing a proprietary system behind the scenes, and this page is where we share the journey.
We publish the problems we encounter and our approaches to solving them — not the implementation details, but the thinking. The challenges on the road to AGI are fascinating and worth discussing openly.
Research Areas
How does a system know what it does not know? We explore approaches to uncertainty estimation and honest confidence calibration.
Storing everything is easy. Knowing what to keep, what to consolidate, and what to discard is the real problem.
Can useful behaviour emerge from simple agents without a central controller? We test the limits of decentralised intelligence.
Reasoning that works at small scale breaks down at large scale. We investigate approaches that degrade gracefully.
Safety bolted on after the fact is fragile. We explore architectures where alignment is structural, not an afterthought.
Language models manipulate symbols. How do you connect those symbols to meaning in the physical world?
Talos is an open research project. If you work in distributed AI, meta-cognition, or autonomous systems, we would like to hear from you.