4 Charts That Show Why AI Progress Is Unlikely to Slow Down
time.com
time.com
These graphs literally don't matter.
>First Graph showing competence
They skyrocket then flatten. Just as I expect to happen with LLMs and AI Art.
>Amount of compute for AI
Probably the only valid graph here. Lots of people are using it now, so I expect further investment. Although we still might be nearing end game for LLMs, now its going to be all about application.
>Algorithm Efficiency and Data Points graphs
Garbage data is the real issue. More isnt necessarily better.
Real AI is a monster compared to ML.
Even simple things a kid would manage still trip them up.
Eg if it takes 5 shirts an hour to dry in the sun how long would 10 shirts take?
I’ve yet to see one get that even remotely right.
Admittedly a bit of a trick question (for a machine) but that to mean means they’re not yet at 100%
Things do seem to be improving quickly, but if the inputs are largely human written content I do wonder if it's feasible to be much better than humans.