I cannot imagine a scenario where it doesn’t drop off.
The massive recent improvements in GPT’s performance are a result of giving the model enormous numbers of parameters and a wealth of training data. That’s it.
Surely this paradigm cannot scale up indefinitely. Moore’s law is moribund.
Are we going to build a supercomputer that encircles the globe just for the purpose of trying to make the biggest LLM we can?
Also, even we could scale indefinitely, there is no reason to suspect that an LLM with hundreds of quadrillions of parameters will somehow magically spontaneously become an AGI.
It’s tempting to think that the line will go up forever, but that just doesn’t square with reality.
Anyone whose response to this is something like “well, maybe all the brain is doing is just the same thing that LLMs do” is fundamentally underestimating the complexity of the human brain by many orders of magnitude.
It is the most complex system in the known universe and we do not understand it at all.
I am generally optimistic about AI, but to me it is the absolute height of delusional hubris to think that superintelligence is likely to somehow “fall out” of a language model, just as soon as we make one large enough and give it enough training data.
To come to such a conclusion reveals a failure to grasp the magnitude of the problem.