Re:Peter Norvig critically reviews AlphaCode's code quality
by nvrmnd on Friday December 16, 2022 @09:51PM
(#34023885)
This is also a very interesting concept/question when it comes to AI writing software for practical applications.
Would a large language model benefit from descriptive variable names when it comes to code understanding, or would this just be for human benefit? My first thought was that a computer (e.g. compiler) does not care what the token is named. But a large language model may actually benefit from it, certainly chatGPT is sensitive to it.
But then, doesn't it mean that it could be fooled by misleading variable names (as a human would), something we would criticize any system for. "Alpha*** is easily fooled by changing variable names, making it completely unusable for blah blah".
Would a large language model benefit from descriptive variable names when it comes to code understanding, or would this just be for human benefit? My first thought was that a computer (e.g. compiler) does not care what the token is named. But a large language model may actually benefit from it, certainly chatGPT is sensitive to it.
But then, doesn't it mean that it could be fooled by misleading variable names (as a human would), something we would criticize any system for. "Alpha*** is easily fooled by changing variable names, making it completely unusable for blah blah".