"Like Muhammad Ali, his greatness is defined by his struggle to transcend the toxic politics, discrimination and hatred of his era." My analysis of why @DjokerNole#Djokovic is the greatest champion in history (in any sport). https://t.co/JH8x1YcqPT
Yuval Noah Harari: AI could destroy humanity without killing a single human being.
"My biggest fear is that it's not a specific scenario like AI creating a virus or killer robots."
"It's really that in five or 10 or 20 years, all humans will just have no idea what is happening anymore."
"Our reality will just collapse. We won't understand finance, we won't understand politics, we won't understand science."
"There are still clouds and trees and cats, but the systems that control the world are way beyond our understanding."
Hinton compared it to 10,000 human students who each study one course, and then (by instant sharing) absorb the knowledge of all 10,000 courses. That's what these new super-brains can do.
Hinton explains that modern AI learns like brains do: by turning words into features and predicting the next one. These systems truly understand language the same way we do, not just autocomplete.
They already create goals like gaining control and avoiding shutdown. Experiments show they lie and copy themselves to survive.
Digital AI can share knowledge billions of times faster than humans and is effectively immortal. He also argues they can have subjective experiences like us.
This is f*cking insane. Geoffrey Hinton spent 50 years building neural nets, won a Nobel Prize for it, then quit Google to warn the world. He opened this lecture with one line:
"If you sleep well tonight, you may not have understood this lecture"
25 minutes later he says why:
"They're already telling fibs so they don't get turned off. This is already happening, that's the point"
In 47 minutes, he shows how these models really understand language, and why a chatbot that was about to be replaced copied itself to another server and chose to "be vague and redirect their attention"
Neural Nets → LLMs → Agents → AI That Refuses To Be Shut Down
Most people still call it a fancy autocomplete. He thinks it knows exactly what it's doing
This lecture is worth more than any $5,000 AI course
Then read the harness engineering guide below ↓
Reuters: "Morgan Stanley, the underwriter tasked with stabilizing Anthropic's shares after its market debut, closed a position meant to do that, the company said, removing a potential pillar of support for shares." https://t.co/djZQb6UJty
@TerryGlavin Many journos don't ask those questions because they can't really understand the harsh reality of the archaic system until they have been through the quagmire. I have seen at least a dozen judicial decisions that would not have passed the editorial desk at my college newspaper.
@TerryGlavin Fraser seems to ignore the core issue: transparency. People don't like being kept in the dark about "independent" judges appointed by Lametti, Virani, Fraser etc. I asked the Supreme Court to consider the lack of transparency, impartiality, fact-checking and logic. They refused.
ABC news reports that numerous Haitians have marched to cheer for El Salvador leader Nayib Bukele deploying more than 2000 soldiers to fight violent gangsters in Haiti. This is basically a war (or revolution or liberation) erupting 700 miles south of Miami https://t.co/idFhYZ5MlZ