4,797 karma · joined April 4, 2022
Then there's Path Canonicalization. This is especially hairy, and only needed in certain circumstances where you need to know when two paths are really on the same filesystem or not (Mount points and symbolic links can cause "C:\dir\file" to be on a different drive than "C:\"). Not helping this is that path canonicalization is broken on Windows in some cases. When you mount a filesystem, you are able to register the volume with the mount manager. If you do register it, path canonicalization will work, but you also get some unwanted side effects (creating a "Recycle Bin" and "System Volume Information" directory). If you don't register it with mount manager, the files are still usable, but Path Canonicalization won't find the drive. You have to enumerate every drive letter to find it, and if you mounted it inside a directory, that's even harder.
"x264vfw" is one tool which supports this format, even inside of an AVI file. So older software that uses Video For Windows can create those files.
For lossy, neither is in a good place, since lossy AVIF beats both lossy JXL and lossy WEBP. Lossless AVIF is a joke and barely competes with PNG.
6502 emulators can easily detect dumb endless loops, and that can turn into what's basically a Z80 HALT.
If you ran a 6502 emulator on a cheap ARM SOC, it would be much lower power.
For the koboldcpp/llama.cpp example, there is a setting where you can change the initial RNG value. It says "seed" in sampler settings. Set it to any constant besides 0 and you get consistent behavior.
That was from when I was testing out someone's claim that koboldcpp got less smart in a newer release, but I was reproducing the exact same output across different versions of the program.