1:03:51 Why even suggest that C++ is too complicated for AI to master? Previous examples showed it was good at generating, explaining and fixing C++ code. Even generating quite complex code involving variadic templates, fold expressions and short-circuit evaluation that novice/intermediate C++ programmers may not be comfortable with (yeah I think we are f*cked).
Did you hear at least once "If programmers get replaced, everyone will"? Well it's true! Yes it's happening to us, but it's also happening EVERYWHERE ELSE, we are not quite there yet, not even web dev which has easier templates to follow has fallen to AI. But one thing is for sure, if everything is not automated once the majority of programmers become redundant, whatever remains, will definitely not last long at all. So don't feel too bad, if your job is done, everyone is also. You see, technology gives us power overall. Think of how AI translators will enable you to communicate in Japan, that's a superpower in and of itself. 20 years ago you definitely needed a translator and if that person was not Japanese, it learned around 4000 letters, each one meaning multiple words in different contexts... SUPER complicated. But these people are in the process of losing their place IN OUR CURRENT CAPITALISTIC SYSTEM. The economic system does not remain the same after a century, and this has been the case for the last 3 centuries (although perhaps to a lesser extent). When everyone is experiencing the same phenomena at roughly the same time, it means the economic system is failing you, not the other way around. Think of this as molting the skin, which happens regularly in snakes and crabs. Technology gives you power, but takes away power in our current OUTDATED system. We need a new one to support the new technologies coming through.
LLMs will increase the number of legacy code, and because they use the available code on the internet for training i see a risk of AI tools generating code using deprecated libraries, and new features like C++23 std::expected will be slower to be integrated in modern codebases because the probabilistic nature of AI models will have more examples of legacy codebases. Maybe in the future a combination of compiler specification, documentation and use of compiler as verification of correct compilation can be used to generate synthetic data to train new AI models with modern features of programing languages, but this looks like a hard problem for the development of programing languages if LLMs are generally adopted.
1:03:51 Why even suggest that C++ is too complicated for AI to master? Previous examples showed it was good at generating, explaining and fixing C++ code.
Even generating quite complex code involving variadic templates, fold expressions and short-circuit evaluation that novice/intermediate C++ programmers may not be comfortable with (yeah I think we are f*cked).
Did you hear at least once "If programmers get replaced, everyone will"? Well it's true! Yes it's happening to us, but it's also happening EVERYWHERE ELSE, we are not quite there yet, not even web dev which has easier templates to follow has fallen to AI. But one thing is for sure, if everything is not automated once the majority of programmers become redundant, whatever remains, will definitely not last long at all. So don't feel too bad, if your job is done, everyone is also.
You see, technology gives us power overall. Think of how AI translators will enable you to communicate in Japan, that's a superpower in and of itself. 20 years ago you definitely needed a translator and if that person was not Japanese, it learned around 4000 letters, each one meaning multiple words in different contexts... SUPER complicated. But these people are in the process of losing their place IN OUR CURRENT CAPITALISTIC SYSTEM.
The economic system does not remain the same after a century, and this has been the case for the last 3 centuries (although perhaps to a lesser extent). When everyone is experiencing the same phenomena at roughly the same time, it means the economic system is failing you, not the other way around. Think of this as molting the skin, which happens regularly in snakes and crabs. Technology gives you power, but takes away power in our current OUTDATED system. We need a new one to support the new technologies coming through.
Everything eventually turns to philosophy! That comment alone got this video a like from me 😇
LLMs will increase the number of legacy code, and because they use the available code on the internet for training i see a risk of AI tools generating code using deprecated libraries, and new features like C++23 std::expected will be slower to be integrated in modern codebases because the probabilistic nature of AI models will have more examples of legacy codebases.
Maybe in the future a combination of compiler specification, documentation and use of compiler as verification of correct compilation can be used to generate synthetic data to train new AI models with modern features of programing languages, but this looks like a hard problem for the development of programing languages if LLMs are generally adopted.
Newer version of this talk (accu 2024): th-cam.com/video/bz7M6rOKZLo/w-d-xo.html