Both. If you’re working on modern C++ projects, most devs around me shun AI because it’s not up to date with the evolution of the language. Even more in niche stuff that is evolving every month (like C++ or other languages hard to learn).
AI because it’s not up to date with the evolution of the language.
😯
May I ask, what kind of niches that are?
Admittedly, I work in industrial automation, where tool changes happen rather on a scale of multiple years rather than months, but I find it hard to imagine projects in which the whole C++ development base changes in less than half year intervals (as that is a typical knowledge-cut-off period for frontline models).
But even than the models will know the working state of the C++ standard proposals, which typically don’t change that much any more during the last few months before official release.
Have they considered supplying updated specs and guidelines via RAG? Or just preloading it into the context?
If you work for startups that deals with any kind of robot, you have at most 10 developers and you cannot waste your time dealing with MCPs, agents, RAGs, or the brand new stuff that changes every week in the AI world. Also they are very reluctant to send their source code to american companies.
Both. If you’re working on modern C++ projects, most devs around me shun AI because it’s not up to date with the evolution of the language. Even more in niche stuff that is evolving every month (like C++ or other languages hard to learn).
😯
May I ask, what kind of niches that are?
Admittedly, I work in industrial automation, where tool changes happen rather on a scale of multiple years rather than months, but I find it hard to imagine projects in which the whole C++ development base changes in less than half year intervals (as that is a typical knowledge-cut-off period for frontline models).
But even than the models will know the working state of the C++ standard proposals, which typically don’t change that much any more during the last few months before official release.
Have they considered supplying updated specs and guidelines via RAG? Or just preloading it into the context?
If you work for startups that deals with any kind of robot, you have at most 10 developers and you cannot waste your time dealing with MCPs, agents, RAGs, or the brand new stuff that changes every week in the AI world. Also they are very reluctant to send their source code to american companies.