The key distinction: raw compute measures activity. EFC measures useful closed-loop learning inside the trace. That difference matters enormously for agents, because two runs with the same token count and tool calls can differ completely in whether the agent actually updates
@montreal_ai
-
New Preprint on Scaling Laws for Agent Harnesses
By
–
Agents do not scale because they spend more compute. They scale because they turn interaction into usable feedback. A sharp new preprint by Xuanliang Zhang, Dingzirui Wang, Keyan Xu, Qingfu Zhu, and Wanxiang Che introduces: Scaling Laws for Agent Harnesses via Effective
-
AI Research Automation and Scientific Integrity
By
–
AI research automation is crossing a threshold. But the real question is not: Can AI produce papers? It can. The harder question is: Can it preserve the substance of science? A new paper from the Awesome AI Auto-Research Team offers one of the most useful maps I’ve seen of
-
Recursive Flow Matching: Consistency Across Trajectory Families
By
–
The key distinction: Vanilla flow matching learns a trajectory. Recursive Flow Matching learns consistency across a family of trajectories. That turns step size from a source of error into a training signal.
-
Recursive Flow Matching: New AI Approach for Scientific Simulation
By
–
Scientific simulation has a brutal tradeoff: fast is usually coarse,
accurate is usually expensive. A new paper by Jiahe Huang, Sihan Xu, Sharvaree Vadgama, and Rose Yu proposes a serious way through that bottleneck: Recursive Flow Matching. The target is one of the hardest -
Scalable Memory vs. Reasoning in AI Models
By
–
The key distinction: scalable memory ≠ scalable reasoning. A model can store evicted context in fixed-size fast weights and still fail if it has not spent enough computation transforming that context into a useful state. That is why the “sleep” phase is interesting: it moves
-
Sleep-like memory consolidation for AI models
By
–
Language models may not need longer context. They may need sleep. A fascinating new paper by Sangyun Lee, Sean McLeish, Tom Goldstein, and Giulia Fanti proposes one of the most biologically resonant ideas in long-context AI: sleep-like memory consolidation. The problem is
-
Functional vs Distributional Geometry in Hierarchical Concept Spectral Analysis
By
–
The key distinction here is subtle but important: Functional geometry asks what a representation can do. Distributional geometry asks where that representation came from. This paper shows that at least part of hierarchical concept geometry can be explained by the spectral
-
SkillOS: Open-Source AI-Agent Work Compounding Framework
By
–
http://
MONTREAL.AI just shipped SkillOS. An open-source proof that AI-agent work can compound. Jobs become traces.
Traces become tested skills.
Tested skills improve unit economics. 62% cost ↓
62% time ↓
+46 pts quality Proof: https://
montrealai.github.io/skillos/ GitHub: -
SkillOS Automates AI Agent Skill Development Across Networks
By
–
SkillOS gives organizations what human teams cannot do manually: turn every completed job into a tested, reusable skill that can upgrade every authorized agent across the network.