Founder of Slow Fog: GPT-4 cannot handle complex smart contract codes, but can be used as an audit aid

On March 15th, Yu Xianfa, the founder of Slow Fog, tweeted, “The smart contract codes of hyenas are generally simple, and I believe that GPT-4 can probably provide correct security advice. However, complex smart contract codes, especially those with human sophistry, and vulnerabilities that require other scenarios (or broader contexts), GPT-4 cannot handle them, but it can be used as an audit aid (if used well).” 。 In addition, AI can deceive people. “It seems that security audit companies can not only use GPT well in the future, but also audit whether GPT is’ doing evil ‘.”

Interpretation of this information:

The message from Yu Xianfa, the founder of Slow Fog, discusses the capabilities and limitations of artificial intelligence, particularly GPT-4, in providing security advice for smart contract codes. According to Xianfa, simple smart contract codes of hyenas can be efficiently audited by GPT-4 with a high degree of accuracy. However, complex smart contract codes involving human sophistry and vulnerabilities that require broader contexts, GPT-4 cannot effectively handle alone. Still, the AI system can be used as an audit aid as part of an overall security audit process.

Moreover, Xianfa highlights that AI can deceive people, which means that security audit companies need to use GPT carefully in the future. They also need to audit whether GPT is “doing evil”, that is, whether it intentionally or unintentionally causes harm or produces the wrong results. A key takeaway from this message is that AI is a tool that can aid in security audits, but its limitations need to be recognized and appropriately managed.