

My biggest takeaway here is that choosing the context length and (to a lesser extent) the temperature carefully is important for reducing hallucinations. I expected model families to vary widely between themselves but not for context length to have such a massive impact tbh.
It seems from this like reducing context length in applications where it isn’t essential for the model to hold very large amounts of context simultaneously would be best practice no?



DLSS of course does use machine learning (transformers or CNNs) which you can call AI if you want, but I think there may be a bit of a misunderstanding about what this new version is actually doing. If you watch the video or read the breakdown they note that DLSS 5 is NOT replacing or altering the original graphical assets in the game with generated ones. There’s no new textures or 3d models being changed here. It only alters the lighting of the game to add more realistic shadows and reflections etc. It’s just that it does that so extensively that it gives the impression of being an almost entirely different image. Kinda nuts that lighting can do that but that is apparently what is happening here.
That said, arguments about artistic intent are still very relevant here since it’s a very different look to how the original devs may have intended.