Pohsuan Wu

[May 2025]short note

  • This might be something you learn sooner or later as you are coding…..even if your space complexity is the same, some methods are still better than the other ones and there is a reason why some method is preferred more than others
    • For example, array is always going to be better then hash map. As hash map are likely built with a larger size array as a base to make sure they got space for all the potential inputs
  • You can write all the code you want to integrate ChatGPT to your project, and it seems like free tier also has some capacity to use. However, it will not work until you put money in.
    • I wonder if it used to be ok, then the pricing chanced without all the documents updated
  • When making a project with LLM in mind, make sure you think about business model since every query costs
    • Suck up! does a pretty good job about it
    • It might be better to use LLM for something more lengthy and fancy, or rewrite lines into different flavor/personality
  • By putting your definition of relationship between variables into code makes things faster as you skip the conversion of A->B (if applicable), but it also limits the possibility of extending the function. Pick whichever that fits the need.
  • Edge AI looks pretty interesting, it should provide a better distribution of LLM model to games without needing to be always online
    • It probably will be with embedded system first
    • Before being part of the game, it is likely to go through a few generations
    • Probably still need a few year done the line for it to be part of PC…but console might be able to have it in the next generation? If this happens, PC game and console game will separate til PC catches up, vise versa
    • If it really is used in game, that means we can no longer have the same-same experience for everyone, but it should be fine as long as it is used for not-lore important elements, just like LLM.

, , , ,

Published by

Po-Hsuan Wu

Leave a comment