New AI agent uses AST for theorem proving, cutting costs

Researchers have developed a new theorem proving agent called AoA (Agent over AST) that operates on the abstract syntax tree (AST) of a programming language, rather than its serialized concrete syntax. This approach significantly reduces token consumption and API costs, making large language model-based proof agents more scalable. AoA achieves this by directly integrating proof operations and states within the AST, leading to faster execution and improved problem-solving capabilities compared to existing agents like Amazon's Isabelle Agent on benchmarks such as miniF2F and NTP4VC-Pearl. AI

IMPACT This research could make LLM-based theorem proving more cost-effective and scalable for program verification and formal mathematics.

RANK_REASON Research paper detailing a novel AI agent for theorem proving. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Isabelle Agent
• Joshua Jun Leang Ong
• miniF2F
• Minilang
• NTP4VC-Pearl

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →