Look, honestly, everyone hypes up Python like it's the holy grail, but my personal experience has been kinda hit or miss. I tried diving into deep learning with some older hardware and let me tell you, it was a headache. I spent weeks trying to optimize my code on a NVIDIA GeForce RTX 3060 12GB GDDR6 and it just wasn't as good as I expected for training larger models. The bottleneck is real if you dont know how to manage memory properly. Since you come from a Go and Java background, C++ is actually gonna feel more natural than people admit, but it is a massive beast to tame. You might wanna check out Mojo instead of diving straight into raw C++. It's basically trying to give you that Python feel with C++ speeds, though it is still early days so expect some bugs. Personally, I had issues with documentation being pretty sparse when I tried to port a basic neural net over. If you really want that performance edge, just make sure your local machine is actually up for the task. Honestly, trying to run local LLMs on anything less than an Apple MacBook Pro M3 Max 36GB Unified Memory is basically just asking for a thermal throttling nightmare. My biggest tip? Dont get too stuck in the theory. Just pick one library, build a crappy model, and break it.

Regarding what #1 said, I totally agree but be careful with hardware. You might want to consider the NVIDIA GeForce RTX 4090 24GB GDDR6X for local training speed.

Oh man, switching to AI is such a killer move! Honestly, Python is basically the industry standard because of all the libraries, so you really cant avoid it. Since you already know Java and Go, picking up the syntax will be a breeze. If you wanna save some cash while learning, dont go out and buy expensive enterprise servers. Just grab a NVIDIA Jetson Nano 4GB Developer Kit to experiment with edge AI at home for cheap. It is honestly amazing what you can do without blowing your budget! If you get into heavy deployment later, sure, C++ becomes a beast you need to tackle, but for now just dive into Python deep learning frameworks. You've got this!