The latest iteration of Tesla’s self-driving software FSD V12, is making waves, and the early reviews are nothing short of astonishing. Straight from the horse’s mouth, renowned machine learning guru James Douma has given Elon Musk’s masterminds a resounding ovation after test-driving this FSD system, full video check out from X @Rebellionaire.
James Douma (@jamesdouma) experiences FSD v12 for the first time.
How does v12 do compared to v11 in tricky spots near Pasadena, CA? pic.twitter.com/uNgXAL8aRx
— Rebellionaire (@Rebellionair3) March 5, 2024
After putting FSD V12 through its paces in the tricky, winding streets of Pasadena, California, Douma’s verdict left no room for doubt. “V12 is going to be over 100x reduction in interventions compared to V11,” he exclaimed, his voice brimming with awe. “This is not an incremental upgrade; it’s a leap forward.”
But what exactly sets this latest iteration apart from its predecessor? According to Douma, the degree to which FSD V12 accurately mimics human driving behaviors is “startling.” Tesla’s ability to capture the good habits while leaving the bad ones behind has exceeded even his lofty expectations for such an early release.
“The range of behaviors that are being effectively implemented with E2E path planning and control suggests to me that they can get the rest of the way to ‘better than human’ solely on top of more and better data and training,” Douma mused, his words carrying the weight of a seasoned expert.
In a testament to the software’s prowess, Douma even went as far as to ponder whether Tesla’s Autopilot team might shift their development approach altogether. After all, when such substantial improvements can be achieved through refining data and training, why reinvent the wheel?
Elon Musk himself chimed in, acknowledging Douma’s “accurate assessment” while revealing that training compute is currently the limiting factor – a hurdle that is “being resolved fast.”