Comments by "" (@d3stinYwOw) on "University Accused of Selling Student Papers for LLM Data" video.

  1. 1. There are nice models which train on input from another LLM, like OpenHermes 2. Chat gipiti went crappy mostly because of jailing it to hell and to get something nice out of it, you need to jailbreak it, just like iOS or Android (aka Root) 3. Academias now have anti-LLM mechanisms in anti-plagiarism software, but there are already AI powered tools which can fool AI-powered tools which combat AI generated data. 4. Issue with copyrighted and copyleft (GPL, AGPL, LGPL) data is only in GPT-3.5/GPT-4, which powers your Github CoPilot, and their latest attempts to provide a 'muzzle' for it, it's only post-generation, their models are already 'spoiled'. What about that topic? ;)
    1