An Empirical Study of AI-based Smart Contract Creation
Abstract
Talk at Berkeley RDI presenting an empirical study evaluating the quality of smart contracts generated by LLMs.
Key points:
- Systematic evaluation of ChatGPT and Google Palm2 for Solidity code generation
- Analysis of validity, correctness, and security of AI-generated smart contracts
- Security bugs introduced by LLMs in generated code
- Impact of input quality and prompting strategies on output
Takeaway: While LLMs can generate functional smart contracts, they introduce security vulnerabilities that require careful review and testing before deployment.
