On a more positive note, LLMs (or their successors) could be used to create a perfect tutor. Taylored for every individual, automatically adjusting learning material difficulty, etc.
But yeah, first we'll go through a few (?) years of the self-defeating "ChatGPT does my homework" and the necessary adjustments of how schools/unis function.
So you suggest training a model for each individual student? Because LLM "inference" sure as hell isn't capable of tailoring anything to anything, or change in any way.
And also, how is personalized bullshit better than generic bullshit? We'd need to solve the bullshit problem in the first place, which is mathematically guaranteed NOT to be possible with these types of architectures.
But yeah, first we'll go through a few (?) years of the self-defeating "ChatGPT does my homework" and the necessary adjustments of how schools/unis function.