🔮 Exponential View #565: Autoresearch; the solar supercycle; an agentic nation; ChatGPT Olympian, seeing fraud & moving asteroids++

2026-03-15 · Source: Exponential View · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

This intelligence brief presents three distinct analyses: the "solar supercycle," the rapid adoption of AI agents in China, and Andrej Karpathy's "autoresearch" project. The solar supercycle posits that solar power, with its decreasing costs, can solve major civilizational problems like water scarcity and carbon capture, with a model predicting market thresholds based on fifty years of Wright's Law data. Concurrently, China is experiencing a surge in AI agent adoption, exemplified by 1,000 people lining up for OpenClaw, with local governments offering multimillion-yuan subsidies and major platforms providing easy access. This contrasts sharply with lower AI optimism in the US. Finally, Karpathy's autoresearch demonstrated an AI agent running 700 experiments in two days, finding 20 improvements that cut the training time for a GPT-2 level language model by 11%, from 2.02 hours to 1.80 hours.

Key takeaway

For AI Scientists focused on model development and optimization, Andrej Karpathy's autoresearch highlights the immediate potential of AI agents to accelerate experimental cycles. You should explore integrating similar automated research frameworks into your workflows to rapidly identify performance improvements, potentially cutting training times and resource consumption for language models and other AI systems.

Key insights

Declining solar costs, rapid AI agent adoption, and AI-driven research automation are reshaping global energy and technology landscapes.

Principles

• Learning curves drive solar cost reductions.
• AI agents can significantly boost developer productivity.
• Automated research accelerates model optimization.

Method

The solar supercycle model uses Wright's Law data to predict market thresholds based on cost reductions. Autoresearch employs an agent to run numerous experiments, retaining only genuine improvements to optimize a target metric.

In practice

• Explore solar.exponentialview.co for energy cost projections.
• Investigate OpenClaw for AI agent deployment.
• Consider automated experimentation for model training optimization.

Topics

• Solar Energy
• AI Agents
• China AI Adoption
• AI Research Automation
• Energy Transition

Code references

• karpathy/autoresearch

Best for: AI Scientist, Entrepreneur, Investor, Research Scientist

Related on AIssential

• Karpathy's Autoresearch On My AI Polymarket Trading Bot — An auto-research loop can autonomously evolve trading strategies for arbitrage bots.
• Andrej Karpathy's new open source 'autoresearch' lets you run hundreds of AI experiments a night — with revolutionary implications — AI agents can autonomously conduct and optimize experiments, accelerating research and discovery across diverse fields.
• 🔮 Exponential View #571: DeepSeek shows the future, again; drones on a learning curve; solar goes up, LLM pixels & tennis robots++ — Rapid iteration and cost-efficiency are critical for technological advantage in both warfare and AI development.
• The Download: coding’s future, the ‘Steroid Olympics,’ and AI-driven science — AI is rapidly transforming coding, scientific research, and societal norms, raising both opportunities and significant concerns.
• Autoresearch, Agent Loops and the Future of Work — AI agentic loops enable autonomous, iterative experimentation and continuous improvement across various domains.
• 😺 🎙️ We're LIVE: Learn how to finally publish that app... — AI agents are reshaping software development and accelerating fusion energy research, demanding new skills and approaches.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Exponential View.