It’s so fun when it’s so specific about some detail with casual confidence that is based on absolutely nothing at all. I know ultimately it’s architecture is more akin to a predictive word generator, but it seems so much better.
Saw a clear demonstration and it is wild that the output is consistent, but at least in the model I saw being run, every word is generated without it having considered what the word after would be or what the general concept it is going for. For a human one has to already know the concept before he/she starts putting words to it, but at least the models I’ve seen explained with detail, it manages to assemble it word by word without knowing where it is trying to go in advance.
It’s so fun when it’s so specific about some detail with casual confidence that is based on absolutely nothing at all. I know ultimately it’s architecture is more akin to a predictive word generator, but it seems so much better.
Saw a clear demonstration and it is wild that the output is consistent, but at least in the model I saw being run, every word is generated without it having considered what the word after would be or what the general concept it is going for. For a human one has to already know the concept before he/she starts putting words to it, but at least the models I’ve seen explained with detail, it manages to assemble it word by word without knowing where it is trying to go in advance.