[ilaksh]

One can hope, but that would be a complete break from the history of computing.

There is still relatively low-hanging fruit because this is a very specific application that suddenly has a huge amount of attention. So there are software improvements, model improvements, and hardware improvements. Probably before we even start a truly new paradigm, Nvidia can get close to 10X by focusing on GPT. Model improvements can likely get another order of magnitude.

There are also radically different compute-in-memory paradigms in the pipeline.

[bitL]

We'll see I guess. All that GPT models showed us so far is that we humans aren't as intelligent as we thought. We might be forced to redefine what makes us human as intelligence might not be our most distinguishing trait fairly soon.

[ilaksh]

Most of the human traits missing are things we share with many animals. Not that that diminishes them.

Things like high-bandwidth senses tied to a body, stream of subjective experience, control seeking, self-centeredness, self-preservation, reproductive instincts, emotions. Also GPT doesn't have certain types of adaptivity yet. Or anything else that is part of being alive.

[kaba0]

Or.. any kind of higher thought and reasoning, which is at best imitated by LLMs.