Colombo paper describes self-organizing AI researcher for optimization theory
The authors present Colombo, an AI system that organizes its own research programmes in oracle-based optimization, pairing algorithm design with adversarial testing and Lean-based certification of its mathematical claims.
In plain words
Researchers describe an AI system that plans and carries out mathematical research on its own. In optimization, it works to find both the best possible methods and the hard limits no method can beat, checking its proofs with formal tools. The authors report strong results in more than fifty problem settings.
What was announced
Colombo, introduced in an arXiv paper, is described by its authors as a self-organizing AI researcher, first applied to oracle-based optimization. In this field, the core question is how efficiently a class of problems can be solved through a fixed information interface, and what costs no admissible algorithm can avoid. The system maintains a Complexity Atlas, which the authors call a living map of the gap between achievable guarantees and unavoidable limits, and it converts unresolved gaps into adaptive research programmes that coordinate conjecture formation, mathematical investigation, and replanning as evidence accumulates.
Two mechanisms drive the work. Adversarial Co-Ascent develops algorithms and hard problem families through reciprocal challenge, which the authors say exposes how information is used or concealed, guiding stronger methods and proofs of unavoidable costs. Independent review and Lean-based formal certification check mathematical claims, while a Long-Term Memory carries supported findings and research experience across settings. The authors report that Colombo has established optimal or near-optimal rates for more than fifty settings in the Atlas, and they frame the system as a step toward Autopoietic Science, a research process whose validated advances regenerate the conditions of its own progress.
What it does not show
- The evidence is the paper's abstract only; it reports the authors' own claims, and no independent evaluation, benchmark details, or public availability of the system are described.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.