You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
This statement is from Jensen Huang at the DealBook Summit today and this is only made possible by AI currently designing AI. TSMC is ruled critical to this effort and there is no second source in the near future.
The AI contest is shaping up to become a contest between Amazon, Microsoft, AMD and Nvidia with links and alliances forming and to be formed. One thing all the participants have in common is TSM.
Addition, Elon Musk feels we are only three years away from AI equaling human intelligence at the same conference.
Any thoughts, comments or observations appreciated.
This immediately reminds me of the annual Elon Musk predictions over the last decade or so that Full Self Driving would be ready by the end of the year. Those predictions were an obvious BS. Huang's prediction might be more credible at least for some types of activities. Whether TSMC is critical here or not is a totally unrelated issue.
AI designing AI sounds like a classic case of doubling down on a failed ideology. We don't even have a theoretical description of intelligence, let alone a processor to implement it. We can correlate stuff using massive amounts of data and create algorithms to recognize and generate patterns, but causal understand and the capacity to build knowledge in realtime are still completely absent. Take the dumbest kid in the class and show him a picture of a cat, and he knows what a cat is for the rest of his life. Different breeds, sizes, species, artistic representations, it doesn't matter. Train the latest transformer on a database of a million cats and it'll still be wrong 10% of the time.
Nvidia CEO Jensen Huang says a major advance in artificial intelligence, making possible machines that can think and learn for themselves, may not be far away.
This immediately reminds me of the annual Elon Musk predictions over the last decade or so that Full Self Driving would be ready by the end of the year. Those predictions were an obvious BS. Huang's prediction might be more credible at least for some types of activities. Whether TSMC is critical here or not is a totally unrelated issue.
I agree. Present day AI technology does well with highly structured problems and information spaces. LLM AI impresses everyone by passing medical board exams, legal bar exams, taking an ACT, writing essays, and summarizing information. Driving on public roads is a much more open-ended and multi-dimensional problem. Assumptions about traffic laws, behavior of other vehicles, behavior of pedestrians and animals, and even the characteristics of an obstacle (Is it a trash bag that fell off a landscaping truck or a hunched-over human?), make the potential list of required behaviors, and therefore the potential choices, huge. We surely aren't even close to cars that are safe when the majority of other vehicles are driven by humans. The only chance for self-driving for many years IMO is to have dedicated roads for them where FSD can be enabled.
100% agree. We will not have self driving cars in mass. its been 10+ years and we are now 10 years behind original predictions LOL. IF they did work, they would cost $300K in mass production. AI is great at summarizing internet data .... and of course it is programmed to be politically correct. This is not intelligence.... this is data mining (which is great). Its a great tool to be used by humans
I am sure an AI bot is planning to have me killed for saying that. Luckily, it will mix up my name with a person with a similar name and they will be killed by accident. Or the car programmed to run me down will have someone put a cone on the hood and it will be permanently disabled closing the street.
