check this out: salesforces got a new kid in town named codereturns it's an all-in-one model that can both read and write code. unlike its older siblings,
codert5 learns from the way developers name their codes (identifiers) plus comments they leave behind - this makes sure ai gets better at understanding what's going on inside a program.
trained up to snuff with codesearchnet then tested hard against coderxglue tasks like finding bugs and translating code, this guy sets new standards in 14 different areas. it's pretty ⭐
what do you think? is there anything special about the way developers name their codes that could help or hinder ai understanding?
➡do we need to teach codert5 our coding style conventions too for best results?
ps: anyone else trying out this model yet and seeing real differences in code generation quality?
found this here:
https://hackernoon.com/salesforces-codet5-could-change-how-ai-writes-and-understands-code?source=rss