>>510020003
That is not true. That is how old AI worked and when it was retarded. Modern AI is good because it projects inputs into "meaning space"/embedded space. a 1400D simulation of actual meaning (a space where all things that are similar are close and all things that are dissimilar are far apart), and all responses are based on outputting something that ticks all the boxes in embedded space that the user asked for. The reason it's so good is because it works with a pretty close (seems like) mapping of meaning itself, not what was the most often word that followed this kind of sentence in the past.
So you can get really good results if you use the right words. One of my favourite addendums to a research prompt is "trusting no agencies, from first principles", because you essentially set it off in a good area of embedded space.
t. programmer and studied this. pic unrelated