How to vibe code in science: early adopters share their tips

· Source: Machine learning : nature.com subject feeds · Field: Science & Research — Artificial Intelligence & Machine Learning, Research Methodology & Innovation, Life Sciences & Biology · Depth: Novice, medium

Summary

The article describes "vibe coding," a conversational technique where users prompt large language models (LLMs) to generate and implement code for various tasks, often without directly inspecting the code. Coined by OpenAI co-founder Andrej Karpathy, this method allows researchers like climate scientist Zeke Hausfather to quickly create complex data visualizations, such as a 3D thermal helix animation showing temperature spiraling upwards, which he could not have coded independently. Specialized LLM-powered tools like GitHub Copilot, Anthropic's Claude Code, and Google's Gemini Code Assist are emerging, with Anthropic's Claude Opus 4.7 achieving 71% accuracy on the Vibe Code Bench. Over 90% of software developers use AI coding assistants monthly, and AI-authored code constitutes over a quarter of customer-facing code. Researchers at institutions like Argonne National Laboratory and the University of Westminster are adopting vibe coding to accelerate data processing, explore new ideas, and even build functional applications like ClawBio, a bioinformatics code library, in significantly reduced timeframes.

Key takeaway

For AI Engineers and researchers seeking to accelerate development and explore novel solutions, embracing vibe coding with LLMs can drastically reduce development time and lower the barrier to entry for complex coding tasks. You should, however, always verify AI-generated code for critical applications, as tools like Claude Opus 4.7 currently achieve around 71% accuracy, necessitating human oversight for reliability and correctness.

Key insights

Vibe coding with LLMs enables rapid code generation and application development through conversational prompts, democratizing complex programming tasks.

Principles

Method

Users provide clarifying prompts to an LLM-powered tool, iteratively refining the output until the desired code or application is achieved, often without direct code inspection.

In practice

Topics

Best for: AI Engineer, Machine Learning Engineer, Computer Vision Engineer, Research Scientist, AI Scientist, Data Scientist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.