alt Hacker NewsI think it’s pretty clear that they are repeating without “comprehension” - both mechanistically (as in there is no facility for comprehension in their formulation) and in the ways they fail. The standard rs in strawberry, should I walk or drive to the car wash, etc all play on the fact that they don’t have any real world model or thoughts against which they can judge their output, as do many of the jailbreaks which basically play on the fact that the model has memorized patterns.
There are people who argue semantics, that we can call the pattern matching that LLMs do “understanding”, or the moronic “how do we know that’s isn’t all we do” but for the normal use of comprehension, LLMs at a fundamental level don’t.
> should I walk or drive to the car wash
there is an entire genre of riddles based on this kind of misdirection that works on humans, such as "A plane crashed on the border or US and Canada. Where do they bury the survivors?"
As with humans, comprehension is not a binary property of the agent - it is a quality that can be present in some situations and absent in others. LLMs may emit correct outputs sometimes because they do comprehend the input, and emit incorrect outputs in other cases when they do not comprehend the input.
In order to show that LLMs can't comprehend, we'd have to show that there are no (or at least very few) situations in which they exhibit comprehension, not show that there are some situations in which they don't.