Yann Lecun is each an enormous in the field of AI, but in addition the fundamental skeptic with the final potential potential of huge language models. He is the fundamental skeptic in terms of huge language models for real artificial intelligence (i.e. real Aga or actual superintelligence). He believes that AI LLM is not going to have the ability to learn true world physics.
Yann Lecun is a French IT specialist considered certainly one of the fathers of latest deep learning. In 2018, he received the Turing Award, often called the “Nobel Prize in Calculations.” He is currently a professor at the University of New York and the fundamental scientist of AI in Meta (formerly Facebook), where he continues research on machine learning algorithms. His work is the basis of today’s AI landscape, affecting technologies corresponding to speech recognition, satellite image evaluation and suggestion systems.
https://www.youtube.com/watch?v=runfgu8kh-4
He believes that the neural networks of huge language models is not going to take the next step. He thinks that Tesla FSD doesn’t solve robotaks. This happens despite 1000’s of Tesla cars driving without drivers from the factory to the loading of the Dock in Fremont each day. Day or night. Rain or glow. There are also tons of of 1000’s of Tesla rides without human drivers up to 85 meters for a real intelligent recall (parking to recall) or shipping the callor to a automobile parking space. He calls and shipments happen each day in the USA, Mexico, Canada and now China.
In June 2025, Tesla Robotaxi’s breakthrough should see tons of of 1000’s of cars without human drivers who don’t drive paid rides in Austin, and then spread around the world to thousands and thousands of cars in 2026.
It took 6 billion miles of driving this point. Mile per minute video. A minute passed for a mile for 60 miles per hour. Two minutes by 30 miles per hour. 6 billion miles of driving data is about 11,000 years of driving data.
Yann Lecun (AI legend, but LLM AI Skeptic) says that the teenager takes 20 hours to learn to lead. However, 12 -year video data trained child to prepare for learning to drive a vehicle. Yann discusses how the child takes 4 years to learn the basics of the world. So he knows that gaining knowledge in the world and the world model takes time before learning to conduct.
This implies that Tesla AI and FSD are 1000-10000 times less efficient than people learning. Although driving learning perfectly (like Robotaxi’s goal greater than 10 times safer than a person) probably takes human 5-10 years. People may never reach ten times safer than the average person. The almost ideal driving standard means 500 times less efficient for artificial intelligence. Let’s assume 1000 times less performance for LLM. ~ 50,000 years of video training data must be enough to train the real world and most real learning of every kind.
This ineffectiveness may not matter. If we collect 100,000 or thousands and thousands of years of video to learn the real world. Then 1500x more calculated than in January 2025 in about two to three years to process it. I believe it’s going to occur around 2027-2029. We use thousands and thousands of cameras to collect data for 2-3 years, which can allow you to train thousands and thousands of years.
I followed the construction of the XAI data center in Memphis. I do know that XAI adds gas turbines that may allow 1.2 GW of power at the end of this 12 months. It shall be in the range of 1 million GPU (NVIDIA B200S). Many higher tokens are coming.
I forecast with the density of a rack, it is feasible to place 3 million tokens to the Memphis XAI constructing. There have to be power supplied thrice to the constructing. It can be possible. This could also be the end of 2027-2028. 30x more systems than applied to Grok 3 (100K H100). Using NVIDIA Ultra or Dojo 4 100x systems higher than H100. Training 3 times longer. 9000x more calculations. Other network improvements, memory, AI models.
This superclaster can have to process 20 quadrilions of video tokens and other data.
Tesla plans to earn 1 million Teslabots in 2027. This 12 months there shall be 5000-10,000. In 2026 it must be 50,000-100,000. It would mean 100,000 years of humanoid data and video for these bots 2026.
Brian Wang is a futuristic leader of thoughts and a preferred scientific blogger with 1 million readers monthly. His blog Nextbigfuture.com ranks first on the science news blog. It includes many destructive technologies and trends, including space, robotics, artificial intelligence, medicine, anti -aging biotechnology and nanotechnology.
Known for identifying the newest technologies, he’s currently a co -founder of the startup and collection of cash for corporations with a high potential stage. He is the head of research on the allocation of deep technological investments and investor an angel in Space Angels.
A frequent speaker in corporations, he was a TEDX speaker, a speaker of the university of peculiarities and a guest during quite a few interviews for radio and podcasts. It is open to public speeches and advises.
