Use of AI-generated content

[Tirlititi]

For your information, I will not handle anything related to mods using AI-generated content or to pull requests using AI-generated code/description in the future concerning Memoria.

The reasons are numerous for discarding any use of these types of algorithms. Here are the most important from my perspective.

• They rely on huge databases that are not built automatically (for their training process). A huge amount of work has to be done, and since it is so big, it has to be extremely low-paid with bad working conditions. Generative algorithms are all but automatic, with at least tens of million people involved in their training world-wide, mostly in the countries of the global south, although there are many clickworkers in european and north american countries as well.
  We end up with lives destroyed by the mere categorisation work needed to build these algorithms (eg. https://www.theguardian.com/technology/2023/aug/02/ai-chatbot-training-human-toll-content-moderator-meta-openai; just think that in order to generate images, people training the algorithms are daily exposed to horrible pictures, to have a correct labelling).

• Because these databases are so big, they are very demanding in resources and ecologically disastrous. There are more and more data centers and power stations dedicated to these algorithms (including re-opening defective nuclear powerplants such as Three Mile Island's).
  LLMs were already an ecological disaster in 2023 (eg. https://arxiv.org/abs/2412.17376) and that problem keeps growing because of investments in inefficient and costly methods (eg. https://huggingface.co/blog/sasha/ai-energy-score-v2).

• More generally, "Artificial Intelligence" is often used as a marketing buzzword. In the case of generative AI, that buzzword is used to hide the fact it steals whatever it sells (images, code, textual answers, musics...) and to be used as an excuse for decisions that would not be justified otherwise.