it's exactly the same as JPEG images being lossy, while you can see image as a whole (and it is enough for 99% of people), you are obviously missing some details
and the more you rely on those details (professional photography, scientific data) the more obvious it is (to the point of image being useless in some cases)
same with LLMs, we are currently testing how far we can go before we seeing obvious issues
and the more you rely on those details (professional photography, scientific data) the more obvious it is (to the point of image being useless in some cases)
same with LLMs, we are currently testing how far we can go before we seeing obvious issues