Capabilities · Concept
Reasoning as Learnable Rule-Based Process
Position paper defining valid/sound reasoning operationally as a learnable rule-based process to restore construct validity of reasoning evals.
Forecast horizon
Connections
Connections · 3
How this node ties into the rest of the map, and the evidence behind each link.
to · enables 4/10
Vero: Repository-Level Formally Verified Code Generation Benchmark Treating reasoning as a sound rule-based process aligns with repository-level formal proof obligations in Vero.
+4 growth from · implements 4/10
DFM Mimir v1: Open 1B Hierarchical Reasoning Model Mimir’s Hierarchical Reasoning Model architecture is an empirical attempt at structured reasoning with open permissible data.
+4 growth to · supports 3/10
Cognitive Alignment: AI Mirroring Human Reasoning Operational rule-based definitions of valid reasoning support efforts to make model reasoning mirror inspectable human standards.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
Task horizon doubling Mar 2025
The length of software tasks AI agents can do autonomously at 50% reliability has doubled about every 7 months — and since 2024 closer to every ~3 months.
METR → Benchmarks saturating Apr 2025
In one year scores rose by 18.8, 48.9 and 67.3 points on MMMU, GPQA and SWE-bench; real-world software solve rate jumped from 4.4% to 71.7%.
Stanford HAI — AI Index 2025 → Autonomous coding 2025–2026
On SWE-bench Verified (500 real GitHub issues), autonomous coding agents reached ~80–86% by late 2025, up from under 50% in early 2025.
Epoch AI →