Huang's core claim is that raw technical problem-solving -- the definition of smart for a century -- turned out to be the easiest part for AI to handle, making it a commodity overnight. What the machine cannot replicate is the ability to sense problems before they appear, which Huang describes as a compression of decades of failed bets, human friction, first-principles thinking, and wisdom -- earned by surviving situations with no instructions. The future rewards what you see when no framework exists, and the people who built identity around memorization are about to work for the people who learned to read the room.
OKAY LADIES...everyone is espousing the "outcome" of AI BUT,
AI IS NOTHING MORE THAN SEQUENTIAL OPERATIONS THAT TAKE PLACE IN A VERY SMALL SPACE OF TIME. Given a scenario, it does a FUNCTIONAL DECOMPOSITION of the "problem" AND starts the process of information gathering. Once that is done then it moves onto the next phase of eliminating the unknowns!!!! SO FORTH AND SO ON!!!!!
THIS is why the "data centers" NEED a huge amount of water to cool the processors, anything else would be just a lil' bit much like ethylene glycol...DEADLY!!!!!
AI needs to be trained in order to apply known knowledge. How do you train for an unprecedented problem?
I'm still unimpressed. The profusion of AI has engendered the term "AI slop" for products that are flawed simulacra, and "hallucinations" for when AI departs from reality. Companies are now finding that their staff reductions are premature, since the AI product has been found full of errors. We even have arguably primitive AI that kills people (737 MAX MCAS software).
We have AI so stupid it has no idea how to articulate the driving bar on locomotive wheels. Or that objects do not pass through each other. Or that writing on a page does not morph in real time. It also had problems providing the correct number of fingers on a hand, or arms on a torso. These are errors that not even 3rd grade students would make.
The thing about a human being is that you do not know what work of genius he will create, if at all. But you have assurance about the mistakes he will not make.
With AI, you do not know what work of slop it will create, nor what kind of unique and pervasive error it will make.
Humans have the potential for genius. AI has the potential for anti-genius.
Huang's core claim is that raw technical problem-solving -- the definition of smart for a century -- turned out to be the easiest part for AI to handle, making it a commodity overnight. What the machine cannot replicate is the ability to sense problems before they appear, which Huang describes as a compression of decades of failed bets, human friction, first-principles thinking, and wisdom -- earned by surviving situations with no instructions. The future rewards what you see when no framework exists, and the people who built identity around memorization are about to work for the people who learned to read the room.
SOURCE: https://x.com/r0ck3t23/status/2075717484875751755 SOURCE (mirror): https://xcancel.com/r0ck3t23/status/2075717484875751755
OKAY LADIES...everyone is espousing the "outcome" of AI BUT,
AI IS NOTHING MORE THAN SEQUENTIAL OPERATIONS THAT TAKE PLACE IN A VERY SMALL SPACE OF TIME. Given a scenario, it does a FUNCTIONAL DECOMPOSITION of the "problem" AND starts the process of information gathering. Once that is done then it moves onto the next phase of eliminating the unknowns!!!! SO FORTH AND SO ON!!!!! THIS is why the "data centers" NEED a huge amount of water to cool the processors, anything else would be just a lil' bit much like ethylene glycol...DEADLY!!!!!
AI cannot turn a wrench, it may be able to tell u what the problem is but if its not software it is fucked
And if the user doesn’t know what the problem is and doesn’t know how to turn a wrench.
AI needs to be trained in order to apply known knowledge. How do you train for an unprecedented problem?
I'm still unimpressed. The profusion of AI has engendered the term "AI slop" for products that are flawed simulacra, and "hallucinations" for when AI departs from reality. Companies are now finding that their staff reductions are premature, since the AI product has been found full of errors. We even have arguably primitive AI that kills people (737 MAX MCAS software).
We have AI so stupid it has no idea how to articulate the driving bar on locomotive wheels. Or that objects do not pass through each other. Or that writing on a page does not morph in real time. It also had problems providing the correct number of fingers on a hand, or arms on a torso. These are errors that not even 3rd grade students would make.
The thing about a human being is that you do not know what work of genius he will create, if at all. But you have assurance about the mistakes he will not make.
With AI, you do not know what work of slop it will create, nor what kind of unique and pervasive error it will make.
Humans have the potential for genius. AI has the potential for anti-genius.