>>200317
Now I don't mean to sound so dismissive, designing a system that could "just" find the next token at this scale was a massive undertaking that required decades of research and the bleeding edge of computer hardware, and that such a system organically models some degree of abstraction is impressive, but that alone doesn't imply intelligence.
>So what? 3 years ago they could not count the R's in strawberry, right now they can generate pictures and videos indistinguishable from reality
Surely you should know better an ignorant neo-retro luddite like moi that the Deep Learning Models used for image generation are an entirely different branch of "AI" than LLMs.
>code better than any human,
Now, what did I just write about programming?
>solve millenium price problems in all the hard sciences, and so on and so forth.
The Millenium Prize Problems are specifically in Mathematics, none of which have been solved by an LLM. Show me what's actually been solved by LLMs. I expect, much like PHP2Shell, for the groundwork to already have been laid down by human scientists and the machine to iterate over some annoying menial calculations to actually dig at the answer. Something which we've already been doing using theorem proof software for decades.
Actually on PHP2Shell and my earlier rant on programming, I read an interview with one of the researchers in which they explain how the LLM automated the work that would've required hundreds of man hours on Valgrind and fucking gdb to get all the offsets and timing right. Fucking Valgrind and gdb? Yeah no shit that'll take hundreds of man hours, you have no fucking abstraction over what you're doing.
>And the rate of improvement is exponential.
S-curve, and we're nearing the asymptote already.