Tesla has the world’s most advanced AI (not OpenAI). Autopilot runs superhuman edge AI on HW4 (only 20–50W — same as human brain) while Nvidia B300/Rubin use 1,500–2,000W+.
AI5 will jump to 800W for more tasks (distributed inference), but efficiency still unmatched.
FSD 14.3 (10x parameters, reasoning model) incoming — “the last big piece” for robotaxis/Optimus.
xAI already has the lowest cost per token in the industry (fastest/cheapest data centers, highest GPU utilization).
Uses Grok + Claude cross-check for balance. Grok 4.2 first with built-in multi-agent truth validation.
Anthropic (Claude) 4x more capital efficient than OpenAI. xAI/OpenAI race still fluid but OpenAI losing lead.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
As fast as I understand FSD and I guess Optimus use pixel based prediction. The human brain uses something much more akin to JEPA where causal relationships are abstracted. Pixel-based prediction has many flaws which is why FSD is still not fully trusted after so many years. Over the longer term I’m betting on JEPA-based approaches, which actually learn the physics and do it extremely rapidly.
I’m sure Yann LeCun would agree.
What are we talking about here? Does Elon have a new transistor design? The amount of information here is minimal.
Okay, from what level does information derive? If it derives from electronic training it differs in scale and level from quantum or (assuming extension as did Einstein and Schrödinger); sub- quantum — that is, a deeper level at which we humans and all creatures sentient (another bugaboo term) derive.
Advanced chips configured with trillions of logic gates and trained on billions of bytes of information, do not compete with neural organization, but rather (appropriately employed) extend their own sentience. In order to understand this, one has to view sentience as a field (something physics has done for a long time) rather than a location. That is, it is not the chip and it is not the neuron, it is instead the relationship such that the chip with its vast training can interact with neural activity, and the result equals sentience. Sentience does not exist as a noun, but rather as a motion and as a construct in motion.
It’s hard to compare the brain to a neural network chip. But the brain might be considered to have a quadrillion synapses, and have compute equivalent to an exaflop. So it might still be considered significantly more powerful and also more energy efficient than our current chips.
Though it’s hard to compare, some scientists have made estimates ranging from 0.01 to 10 exaflops. Which is a pretty wide range. Synapses are easier to count, but those estimates range from 0.1 to 1 quadrillion. There’s also important information in the topology of which neurons connect to which, and even where on the axon each dendrite connects. Some people have even suggested there is both memory and processing inside the neuron itself, beyond what a “neuron” in an artificial neural network does.
So it’s hard to compare. But I think it’s safe to say the brain is still ahead of the chips, in raw computation and efficiency.