|
|
|
|
|
|
by grey-area
9 days ago
|
|
|
It generates text from a prompt and weights. This is not an oversimplification, this is what it does. It doesn’t know what is good and what is not or a quality holiday for person x is. Humans do not in fact do that, they reason based on a mix of past experience and emotion, consider what is good and what is not and then answer. These are completely different processes. The difference becomes apparent when an LlM makes a mistake for example and then apologises obsequiously and repeats the mistake, or apologises and makes a different mistake. Or when they fail to count letters (one of many flaws monkey-patched by calling tools). They don’t reason they don’t evaluate and they can’t count. This is so so far from human intelligence. |
|
|