There is a common paradox in robotics called Moravexs Paradox, where to train tasks humans find easier (e.g. stirring oatmeal until it’s perfect consistency) - it is very compute intensive and to train tasks that humans find complex (e.g planning, optimizations..) - it is less compute intensive.
This along with algorithmic efficiencies like real time task clustering can unlock beyond highly specific tasks like EV, military drones and so on.
It could make significant progression on use cases where robots emulate or directly replace human labor.
This is a great read on FPGAs! From my experience working on AI inference in autonomous vehicle systems, I agree that FPGAs at the edge are going to accelerate. AI models and system requirements are evolving too fast for teams to over commit to ASICs. I also think the talent barrier is shrinking with universities increasing FPGA education and the popularity increasing driving more talented engineers into the field. Long term, I am bullish on FPGAs continuing to fill the gap between GPUs and ASICs and AMD/Xilinx leading the pack with now around 60% market share and growing.
This could go beyond AI, into robotics as well.
There is a common paradox in robotics called Moravexs Paradox, where to train tasks humans find easier (e.g. stirring oatmeal until it’s perfect consistency) - it is very compute intensive and to train tasks that humans find complex (e.g planning, optimizations..) - it is less compute intensive.
This along with algorithmic efficiencies like real time task clustering can unlock beyond highly specific tasks like EV, military drones and so on.
It could make significant progression on use cases where robots emulate or directly replace human labor.
interesting write up will have to continue to review
Believe me, you’re missing out if you haven’t joined this substack yet @RINGLEADERS HUB
This is a great read on FPGAs! From my experience working on AI inference in autonomous vehicle systems, I agree that FPGAs at the edge are going to accelerate. AI models and system requirements are evolving too fast for teams to over commit to ASICs. I also think the talent barrier is shrinking with universities increasing FPGA education and the popularity increasing driving more talented engineers into the field. Long term, I am bullish on FPGAs continuing to fill the gap between GPUs and ASICs and AMD/Xilinx leading the pack with now around 60% market share and growing.