DigiNews

Tech Watch by Johan Denoyer

← Back to articles

Epicycles All The Way Down (2025)

Quality: 8/10 Relevance: 9/10

Summary

The essay argues that LLMs primarily learn patterns rather than underlying generators, explaining both their impressive capabilities and their failure modes. It surveys the role of reasoning, consciousness, and alignment, and contends that scalable data and architectural innovations may push AI toward more robust, though not fully human-like, understanding. The piece also discusses how memory, tool use, and co-evolution with human systems influence alignment challenges.

🚀 Service construit par Johan Denoyer