AI Writing Code and Conducting Automated Scientific Experiments
AI can now write its own code and perform scientific experiments.
1The AI Revolution
Since 2022, advancements in AI technologies, particularly the emergence of ChatGPT, have marked a turning point. Machines can now not only write code but also solve complex mathematical problems and automate cyberattacks. This transformation is reshaping the global economy.
2Coding Capability
Modern coding agents are fundamentally changing traditional programming methods. Tools like Claude Code not only write infrastructure code but also propose hypotheses, draw conclusions, and plan next experiments. This is revolutionizing the nature of software engineering.
3Automated Experiment System
Tools like the /experiment command automate scientific research processes. They create folders, write Python files, and analyze results autonomously. This saves researchers significant time and boosts efficiency.
4Hyperparameter Optimization
AI can automatically optimize machine learning model hyperparameters. Using techniques like maximal update parameterization, it finds the best settings to enhance model performance, far more efficiently than traditional methods.
5Sequential Experiments
AI agents can conduct a series of sequential experiments, analyzing results and planning next steps. They evaluate findings after each experiment to generate new hypotheses, accelerating the scientific discovery process.
6Unconstrained Search Space
Unlike traditional hyperparameter optimization systems, modern coding agents can modify the code itself. This removes constraints on the search space and enables more creative solutions. AI can assess result consistency and generate theories.
7Automated Scientists
Coding agents, combined with computer use, are evolving into automated scientists. Beyond software engineering, they become general-purpose thinking machines capable of solving short digital problems, shaping the future of research.
8Mathematical Proofs
AI can prove unsolved mathematical problems without human intervention. It generates new proofs for complex topics like Erdos problems, heralding a new era in mathematical research.
9Self-Improvement
AI agents can speed up their own CUDA kernels to run faster. This capability leads to self-improving systems but also raises ethical and security concerns that must be carefully addressed.
10Debugging and Problem Solving
AI excels at debugging and problem-solving, enabling persistent pursuit of goals. Coding REPL agents are rapidly adopted due to their relentless determination and effective search capabilities.
11Golden Age of Computer Science
All computer science problems are becoming solvable. Significant progress is being made in areas once considered challenging, like Go, protein folding, and media generation, potentially marking the start of a new golden age.
12Vision of the Future
AI's ability to write code and conduct experiments offers immense potential for humanity. However, ethical and security considerations must be carefully managed. In the future, scientists and engineers collaborating with AI could achieve even greater breakthroughs.
Original article: Eric Jang — Yapay Zeka ile Kod Yazma ve Otomatik Bilimsel Deneyler