Key facts
- Anthropic is using human software engineers to improve its Claude Code AI tool.
- The project, codenamed 'Marlin', aims to train Claude Code to mimic professional developer skills.
- Contractors are paid up to $280 per hour for prompt creation and code review tasks.
- The process involves A/B testing code from different models to ensure simplified, maintainable output.
- Snorkel AI, a data labeling startup, is managing this work for Anthropic.
As AI models become more sophisticated, the need for specialized human input to refine their capabilities grows. Projects like Anthropic's Claude Code improvement highlight the increasing reliance on human expertise to train AI for complex tasks, moving beyond general data labeling to specialized fields like software engineering. This trend underscores the evolving landscape of AI development, where human oversight and domain-specific knowledge are crucial for creating reliable and high-performing AI systems.