Meta's semi-formal reasoning technique improves LLM code review accuracy to 93% by requiring AI to explicitly trace execution paths before answering

Meta researchers introduced 'semi-formal reasoning,' a structured prompting method that forces LLMs to document logical certificates with premises and execution paths

The technique eliminates the need for expensive dynamic execution sandboxes by using AI reasoning instead of actual code execution

Structured prompting significantly reduces LLM hallucinations and unsupported guesses in coding tasks by requiring systematic evidence gathering

This approach enables AI agents to handle repository-scale tasks like bug detection, patch verification, and code review more accurately and cost-effectively