Reading Ben Recht’s blogpost “Too Much Information”. He argued that as a young scholar, we should choose to do research with strong story, the underlying logic, motivation and insights.

The comments point to two resources:

• A paper named “A measure of research taste” and talk slides “Non Serviam” means “绝不服从” in Chinese.
• The paper propose a new citation-based metric that rewards both productivity and taste. So what is the definition of taste in their paper, they said that “We use ‘taste’ to refer to researcher’s consistency in producing high impact contributions, as quantified by citations”.
  • CAP measure proposed by them is defined as:
    • $\text{CAP} = \sum_{i=1}^P \mathbb{I} [C_i - A_i - P \geq 0]$
    • where $C_i$ is the citation of paper $i$, and $A_i$ is the number of authors of that paper, $P$ is the paper count during the period of interest.
• The talk slides also adds to the discussion some more objective viewpoints. It points out the pros and cons of this kind of research atmosphere (image blow).

Hyung Won Chung’s talk at Stanford CS 25, the slides are available.

• Insightful for making encoder-decoder Transformer to Decoder-only GPT, do the subtraction to thinking about the decision choice of inductive bias.