Re:AI will divide the best from the rest
by archagon on Friday February 14, 2025 @02:54PM
(#43052346)
> Jeff Dean just said 25% of the code being checked into Google's internal codebase is now AI generated.
This assertion feels misleading. When I worked at a FAANG, 10% of our codebase was generated Protobuf. Obviously, this was highly repetitive boilerplate that was equivalent to maybe a few thousand actual lines of code. What does the 25% of LLM-generated code actually look like? I'm guessing it's doing nothing particularly novel or interesting. (Also, I have to wonder if a large chunk of that 25% could be encapsulated as a more reliable and deterministic abstraction.)
This assertion feels misleading. When I worked at a FAANG, 10% of our codebase was generated Protobuf. Obviously, this was highly repetitive boilerplate that was equivalent to maybe a few thousand actual lines of code. What does the 25% of LLM-generated code actually look like? I'm guessing it's doing nothing particularly novel or interesting. (Also, I have to wonder if a large chunk of that 25% could be encapsulated as a more reliable and deterministic abstraction.)