AlphaCode | 11D: Map system input attributes to system outputs

yellow header bar

AlphaCode, a code generator by DeepMind makes clear why the system did what it did (Guideline 11) by highlighting parts of the problem description that the model attended to for generating selected tokens in the code solution (Pattern 11D).

The techniques used in this example have the potential to foster appropriate reliance on AI, by reducing overreliance. Keep in mind that overreliance mitigations can backfire. Be sure to test such mitigations in context, with your AI system’s actual users. Learn more about overreliance on AI and appropriate reliance on generative AI.

AlphaCode attention visualization interface showing a coding problem description and generated solution. Selecting the token const in solution highlighted tokens in the problem based on model’s attention to them.
Image captured February 2024

Related patterns

Submit a pattern
Guideline 11 > Pattern 11D

Pattern 11D: Map system input attributes to system outputs