The tutorials and walkthrough guides for all of that stuff are infuriating.
What do you mean I just import and run “unbelievably_specific_tool” and run it? How am I supposed to do anything other than process this example data with that?
Exactly, coming from Java it was such a big difference. Previously I could click on any class or method and view the implementation, with Python, suddenly everything was a convoluted jungle of methods popping into existence during runtime or being loaded from some opaque binary.
Not enough. It’s why they all need so many fucking resources and why you need to spend $1500 on a video card for a game that could run on a 15 year old GeForce if the developers put any effort into optimizing anything.
Expecting even a medium to large team to waste resources on essentially reinventing the wheel would probably result in shittier games, IMO.
And with their new engine no one is stopping them from adding dlss the entire way, so performance never once has to be a consideration (minus on their test rig of 2 5090s, 128GB of ddr5, and the latest threadripper of course).
You know the majority of LLM stacks are in python, right?
We already live in that world, basically.
Python LLM training produces models that are then refined into things that 3D engines plug into to do AI upscaling and framegen, and then different python trained LLM models use more python to interface with MCPs to write more slop extensions and frameworks in/for game engines, and other programs.
Still different python derived LLMs do image/facial/3d object recognition, get plugged into moar python, and then produce models… often for game engines.
Go ahead, write your 3D engine in Python. I dare you.
You’d write the code in something else and then be like
Aah yes, the entire numerical python stack being carried on Fortran and C’s shoulders
The tutorials and walkthrough guides for all of that stuff are infuriating.
What do you mean I just import and run “unbelievably_specific_tool” and run it? How am I supposed to do anything other than process this example data with that?
Exactly, coming from Java it was such a big difference. Previously I could click on any class or method and view the implementation, with Python, suddenly everything was a convoluted jungle of methods popping into existence during runtime or being loaded from some opaque binary.
Never underestimate my self-loathing and my desire to write the most unusable piece of shit.
How many games write their own 3D engine anymore
Not enough. It’s why they all need so many fucking resources and why you need to spend $1500 on a video card for a game that could run on a 15 year old GeForce if the developers put any effort into optimizing anything.
Expecting even a medium to large team to waste resources on essentially reinventing the wheel would probably result in shittier games, IMO.
And with their new engine no one is stopping them from adding dlss the entire way, so performance never once has to be a consideration (minus on their test rig of 2 5090s, 128GB of ddr5, and the latest threadripper of course).
Screams in GIL
3D engine in LOLCODE or bust
You know the majority of LLM stacks are in python, right?
We already live in that world, basically.
Python LLM training produces models that are then refined into things that 3D engines plug into to do AI upscaling and framegen, and then different python trained LLM models use more python to interface with MCPs to write more slop extensions and frameworks in/for game engines, and other programs.
Still different python derived LLMs do image/facial/3d object recognition, get plugged into moar python, and then produce models… often for game engines.
Its nearly literally an ouroboros.
It’s funny that you think Python is anything else than a wrapper for real languages that get shit done.
Let me know how many LLMs you can run locally without multiple entire suites of python libraries.
Muchless train an LLM.
Yeah, load bearing ‘wrappers’.