Quoting Karel D'Oosterlinck

2026-02-05 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Karel D'Oosterlinck, a researcher at OpenAI, utilized Codex to automate significant portions of his experimental research, incurring a cost of $10,000. Codex performed extensive due diligence for one-off experiments in unfamiliar codebases. This involved exploring relevant Slack channels, reading related discussions, fetching experimental branches, and cherry-picking useful changes. The AI then summarized its findings in detailed notes, complete with links to original sources, and subsequently wired the experiment, making hyperparameter decisions that would otherwise require substantial manual effort from the researcher.

Key takeaway

For AI Scientists and Research Scientists looking to accelerate one-off experiments in unfamiliar codebases, consider integrating AI tools like Codex for automated due diligence and experiment setup. This approach can significantly reduce the manual effort required for information gathering and hyperparameter selection, allowing you to focus on core research questions rather than setup complexities.

Key insights

AI can automate research due diligence and experiment setup, saving significant manual effort.

Principles

• AI can synthesize distributed information.
• Automated decision-making for hyperparameters is feasible.

Method

Codex explores communication channels and code repositories, summarizes findings with source links, then configures and wires experiments, including hyperparameter selection.

In practice

• Automate code exploration for new projects.
• Use AI for hyperparameter tuning.

Topics

• Codex
• Research Automation
• Experiment Setup
• Hyperparameter Optimization
• Codebase Exploration

Best for: AI Scientist, Research Scientist, AI Researcher, Machine Learning Engineer, AI Engineer

Related on AIssential

• 🔬 Automating Science: World Models, Scientific Taste, Agent Loops — Andrew White — AI agents like Cosmos are automating scientific discovery by integrating literature, data analysis, and experimental feedback loops.
• Improving the academic workflow: Introducing two AI agents for better figures and peer review — AI agents can automate complex academic tasks like figure generation and peer review, improving efficiency and quality.
• Rakuten fixes issues twice as fast with Codex — Integrating AI coding agents like Codex significantly boosts development speed, enhances safety, and fosters autonomous project execution.
• How Far Are We From True Auto-Research? — Current auto-research agents generate polished papers but critically lack experimental rigor and result integrity.
• Towards Generalist Agents for Accelerating Scientific Discovery171 — LLMs can accelerate scientific discovery by generating and refining hypotheses within agentic, iterative workflows.
• Towards the realization of AI research by AI: AI Scientist paper published in Nature journal — AI Scientist autonomously generates research papers, passes peer review, and reveals a scaling law for scientific quality.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.