I have found myself wondering a lot lately about the alternative universe where all the effort and resources devoted to developing and using AI coding and debugging agents were instead simply dedicated to fuzzing human-generated code. How would the code quality and total usefulness of the results have compared?
It should have been possible to build a new LLVM alternative, a new chromium like browser and some other similar things imo and that would have been definitely more impactful for me.
It is unimaginable how many static analysis tools or fuzzers could have been created with a fraction of LLM investment too.
For example honggfuzz was created by a single engineer possibly in less than a years’s worth time