Does Microsoft Middle Manager 2.0 stop working in the presence of traffic cones?
More seriously, the distance between "sort of works" and "works" might not be infinite, but it most likely involves fundamentally unpredictable future developments of the current technology. There is no straight line of incremental improvements that gets us there.
It's fairly straightforward to imagine that if you have a 4.77mhz CPU and 64kb of RAM you will soon have a 3ghz CPU and 64gb RAM.
Bigger number going brrr is no guarantee of anything, here, so much so that models using a fraction of GPT4's resources are somewhat competitive.
By all means continue developing the technology, but claims that we are within arms' reach of X for disparate values of X are not exactly supported by anything.
> Does Microsoft Middle Manager 2.0 stop working in the presence of traffic cones?
I think if you put a traffic cone on my IRL managers head she’d stop working too. Maybe a bit different… but maybe not.
I don’t think the poster was saying we’re within arms reach of it, but that there’s a path that takes us from here to there. A MSFT AI manager obviously wouldn’t behave exactly like a real human, but a tool that aggregates and summarizes information from many reports for a high level manager, and helps negotiate priorities is something potentially doable with some prompt engineering and advancements in models.
I respect many of my past managers, and some of them were great mentors and materially improved my life but 50+% of the value managers provide to an organization can be done with a bit of glue between GPT-x and Jira. It’d free up a lot of their time for the remaining 50% too.
More seriously, the distance between "sort of works" and "works" might not be infinite, but it most likely involves fundamentally unpredictable future developments of the current technology. There is no straight line of incremental improvements that gets us there.
It's fairly straightforward to imagine that if you have a 4.77mhz CPU and 64kb of RAM you will soon have a 3ghz CPU and 64gb RAM.
Bigger number going brrr is no guarantee of anything, here, so much so that models using a fraction of GPT4's resources are somewhat competitive.
By all means continue developing the technology, but claims that we are within arms' reach of X for disparate values of X are not exactly supported by anything.