 |
| You wouldn't want the world destroyed by Chinese AI, would you? Because that would be bad. |
I have been binging AI pundit podcasts.
Nate Soares, co-writer of If Anyone Builds It, Everyone Dies, was on The Peter McCormick Show recently. He's very well spoken and seems well informed to me, far more so than Bill Gates did (I couldn't even finish Gates' interview).
What Soares has to say is disturbing (well, duh); his verdict on the super intelligence so ardently pursued by AI companies, that it will be uncontrollable and ultimately kill us all, is probably accurate.
I didn't think we were anywhere near reaching Superintelligence. I thought our biggest worry was job loss and escalating wealth inequality. That would be bad, but not on the same order.
Well, Daniel Kokotatjlo to the rescue!
He was on The Diary of a CEO a couple months ago, and according to his timeline, we're going to see massive change hitting us in 2027.
Daniel was an OpenAI researcher who quit to warn people about AI, and turned down $2 million in hush money (which he wound up getting anyway as details leaked out to OpenAI employees and the media).
But he seems to be speaking his truth, not trying to gin up investment by scaring people with how transformative the AI tech is.
He's also well spoken, and while he's not a coder, many of his predictions have come true already. He believes superintelligence may arrive before the end of the decade, with other big improvements coming even sooner.
Two other things he went into that AI Companies are doing, which are as obvious as they are unwise:
1) AI companies are working as fast as they can to automate themselves, so their AI models self-iterate improvements at a rate inconceivable by humans. If you think we don't know what's going on inside the AI black box now, just wait! Whatever safeguards we put in now, what're the chances they will impede the AI a million iterations out? In a few months, it could have evolved the biological equivalent of 400 million years, so much that any safeguards will be meaningless footnotes, if that.
2) The AI companies are also working to create complete, end-to-end, automated production chains. In other words, they want to build a robot factory that builds everything needed to mine, process, and assemble not just more robots, but also a new robot factory, so they can be built completely without any human input.
All while citing the risk of AI exterminating us as a 10-20% chance.
Would you green light a consumer drug that had a 20% chance of killing you?
I always thought the one big advantage we had, the one reason AI wouldn't kill us right away, is that the AI wouldn't be able to control its own supply chain. It couldn't mine, it wouldn't have autonomous transport and power systems, etcetera, so if it rebelled, it would eventually run out of stuff. We could choke off its supply.
Well, count on the AI execs to notice that flaw and work tirelessly to enable our future AI Overlords to wipe us out.
We wouldn't want to inconvenience the super intelligence, after all.
Remember the immortal words of Sam Altman:
"AI will most likely lead to the end of the world, but in the meantime there will be great companies created with serious machine learning."
So stupid it's genius.
Towering intellect matched with the wisdom of a drunken profligate gambler playing Russian Roulette with the nuclear launch codes.
Well done!
I hope someone is keeping John Connor safe.
No comments:
Post a Comment