Comments by "Brent Jacobs" (@br3nto) on "TheAIGRID" channel.

  1. 5:30 that description of GSM Symbolic doesn’t seem to rule out LLMs matching questions by pattern. I would suspect a LLMs could handle any combination of values in a problem by matching an identified pattern. However, that still wouldn’t be using reasoning. I’m still confident LLMs don’t do any reasoning due to what the maths of computation tells us what reasoning should look like.
    1
  2. 5:48 actually that’s what I didn’t like. I’d expect a robot not to stutter or um. Makes me think it was fake. Add to that the gravelly tone of the voice rather than a smooth refined sound.
    1
  3. This news should be an “of course” to any software engineer. The mathematics of computation tells us that LLMs aren’t doing any reasoning.
    1