Comments by "TheEVEInspiration" (@TheEVEInspiration) on "Apple DROPS AI BOMBSHELL: LLMS CANNOT Reason" video.

  1. A sum of people even :)
    1
  2. I suspect special meaning gets assigned to certain oft used names in training data (think: Alice and Bob in quantum physics). And as the models are not performing intelligent reading, they answer a question never asked. And obviously, these models are just pattern-matching and using statistical probability to assign the next token to the output. It is already surprising they perform as they do IMO.
    1
  3. If they cannot learn to reason, then they are always stuck at the level of the training data, which means its full of mistakes and bias by default. And they cannot self-improve/correct. We could also see this as a safety feature, avoiding AI overlords.
    1
  4.  @jeffsteyn7174  Cope on.
    1