State of AI vs Human Code Generation

The Software Delivery Notebook

State of AI vs Human Code Generation

Play Episode Pause Episode

Mute/Unmute Episode Rewind 10 Seconds 1x Fast Forward 30 seconds

00:00 / 00:14:29

RSS Feed

Share

Link

Embed

CodeRabbit looked at 470 open-source pull requests and produced the State of AI vs Human Code Generation report with their findings.

Crucially, AI makes 1.7x more errors, increasing major and critical defects. The AI-authored/assisted changes cause problems with logic errors, security vulnerabilities, and performance regressions (whereas humans are more likely to make spelling mistakes or have testability issues).

That means the AI is introducing foundational risk that a competent developer should be catching very early. Is the training data just poisoning the model with old insecure patterns?