Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents
ArXiv: 2505.22954v3 | Category: AI Evolution
The DGM represents a step toward AI systems that can build upon their own prior innovations and improve recursively. It outperforms baselines without open-ended exploration by accumulating an archive of stepping stones.
Externalization in LLM Agents
ArXiv: 2604.08224 | Category: Infrastructure
Analyzes the trajectory from Weights (2022) to Context (2024) to Harness (2026). Argues that agent capabilities are now increasingly properties of the broader infrastructure (Harness Engineering).
On the Limits of Self-Improving in LLMs
ArXiv: 2601.05280v2 | Category: Singularity Theory
Hypothesizes that symbolic model synthesis is the "missing link" for a positive feedback loop in RSI. Claims the singularity is not near without this synthesis layer.
X Signal: Jack Clark's RSI Forecast
Source: X/Twitter | Sentiment: Bullish
Jack Clark (Anthropic co-founder) estimates a 60% probability of RSI achieving critical mass by 2028. Signals a shift from lab experiments to industrial-scale self-improvement loops.