This article completely ignores the actual state of the industry, and it seems to be a clickbait trolling the uninformed. For handy reference, in this last year alone:
- AI folded ALL human proteins, a feat we thought would take supercomputers a generation, and now it's being turned into a pharma company to supply tailored drugs. Approval takes a few years, so chill, the advancement already happened and is on its way.
- AI now learns from MINUSCULE amounts of training-data, e.g. an hour of SHAKEY video of someone playing basketball is enough to render a high-res 3D model of THAT person, which you can control with a PlayStation, and its FEET DON'T SLIDE, its LIMBS DON'T COLLIDE... and it can move accurately in numerous ways that were NOT in the training video!
- AI now renders scenes for CGI that are hyper-realistic deformation and damage, specular highlights, soap bubbles, elasticity without collisions, etc. at THOUSANDS of times the speed of our hard-coded algorithms from just two years prior.
-AI can now look at data-sets from scientific experiments, and quickly determine what the physical laws must be; this allows generalization and extraction of symbolic formulae.
- AI is slowly chewing through the LAST few percent of error in numerous measures... which the author seems to think makes them irrelevant? In reality, those image-recognition and speak-synthesis bots are NOT QUITE good enough for products - and that is WHY researchers are struggling for those last few percentage points of performance!! For example, if your AI drops from 5% error to 2.5%, you've HALVED the number of car-driver interventions or amazon warehouse fumbles. That's the difference between "we spent millions on a system that breaks too often to be useful" compared to "yeah, we fired almost everyone, and we're not looking back!" Obviously, that last hurdle is essential BEFORE you'll see products.
Anyone who tracks the actual work being done, from proofs of equivalence between infinite networks and tangent kernels, or re-arrangement of Transformers' matrices to get linear-time, to the lottery ticket hypothesis, knows that AI is still making ALARMINGLY fast progress along more fronts than we expected. Buckle-up; things are just getting started. :)
