Full Self Driving Power

A human brain is the hungriest body organ - responsible for more than 20% of the total energy consumption. A huge drain on the organism.

ChstGPT-4 drains 1 kWh to answer 3 queries (it is estimated to use 300-watt-hours per request, based on the "single-digits cents" cost stated by Sam Altman.

A typical EV battery holds 60kWh. That would be sufficient to handle 200 queries. Text LLM queries.

Compare that to the real time analysis of video streams from 8 cameras (as used in Tesla configuration) to draw driving decisions. It all just does not add up. It seems an AI-powered self driving car computer would drain the battery far more than the car's powertrain. Of course there is stil the Moore's law progress, shrinking geometries etc., but it would be foolish to assume the FSD is just an app update away. The hardware required t run this is absolutely massive. You cannot run FSD on an iPhone.

It also seems the AI model training dataset requirements have outgrown the budgets of even the biggest players in the field. Self-Driving Cars Are Being Put on a Data Diet. It is easy to make this mental leap assuming FSD is "just a piece of software". But the reality is the processing hardware, the storage, and power consumption requirements are still physically unreachable.