[Takeaways] Invited Talk by Junyoung Chung (@Deepmind)
Events

[Takeaways] Invited Talk by Junyoung Chung (@Deepmind)

AI for Code: The Rise of Code LLMs and AlphaCode

  1. AlphaCode is trained based on the "code competition dataset".
    1. Why use the code competition dataset? Because the evaluation metric for the output is clear, it can provide accurate feedback to the model.
  2. To communicate with other models, the direction of code synthesis should be considered.
  3. Originally, large models are created through several stages of development.