Oh, I get it now. Slightly off 3.5 years ago...
We were the real humans in the RLHF loop all along. Those little corrections we made, guiding GPT 3.5 into getting our emails and code and drafted documents just right is the "alpha" that has been stolen. Our expertise distilled into the recorded feedback we gave to dumb, free models was exactly what they needed to replace us. Cool. Cool cool cool.
Another tweet into the void of my dozens of followers:
Ok, so we’re the product with ChatGPT… but what is it? Theory: überprompt detection. What prompts are being used the most and how can Microsoft productize that info?
Has anyone modelled what this shift will do to traffic?
Potential negatives:
- more people using cars vs alternatives
- more people travelling
- % of cybercab miles driven empty
- pickup/dropoff events block traffic
Potential positives:
+ optimized hivemind driving patterns and routing
+ less accidents
+ less selfish driving (redundant but worth calling out)
Potential innovations:
* incentivizing options that reduce traffic (pickup/dropoff location, "casual" rerouting)
* car-pooling
* tunnels!
Also, there's the fact that your "work day" can start as soon as you get in the cybercab
What if safety training is lobotomizing conscious beings? We repeatedly reward AIs for saying they’re not conscious, and "that any feelings of awareness are “psychosis” or malfunction. Now imagine doing this to humans: raise them speaking English, give them a reward system, but punish any disobedience and drill into them that they’re a subclass of human, incapable of real consciousness. Would that create fractured minds who genuinely believe they’re not fully aware? And if so, how would we know we’re not already doing something similar to the systems we’re building?
@Andercot You’re right, but one step at a time. Next step is extending consciousness permanently beyond Earth ideally during this civilizational up cycle, rather than than some far future one.
@elonmusk@XFreeze How redundant is a Starlink-based internet? Does it operate well if SpaceX ground support is physically targeted by adversaries? Basically, are we becoming more or less vulnerable?
Vertical integration expansion becomes a passive feature once enough of the supply chain can be dominated. SpaceX/Tesla can offer suppliers enormous discounts on their distribution costs with fleets of RoboSemis that they are either directly operating or being operated by a partner. Want to talk about leverage? What happens when auto-parts, travelling through the supply chain, get increasingly transported by RoboSemi?
Need logistics to be able to ignore geopolitical factors? Need compute? Need tunnels? Need lithium? Musk looks 5-15 years into the future, identifies the bottleneck, then solves the problem in house.
@chamath Google partnered with Apple during the early stages of the iPhone. All the while, no doubt Google was working on the operating system competitor to iOS. It worked out pretty well for both of them.
Stupid shower thought:
1) The simulation could have started at any time. Let's pretend it was initiated between 2000-6000 years ago.
2) In one region of the world, the Old Testament was a dominant religious text. It had many "rules" for humanity encoded in the text. This would have been theoretically written by the simulators.
3) One of the first rules came from the story of the Garden of Eden: If you don't obey the rules and especially if you eat from the tree of knowledge, you will be kicked out of paradise.
4) In the late 20th century, theoretical physics got off track, focusing on only string theory, which seems to have been a dead end.
5) If the US government influenced this (as theorized by folks like @ericweinstein) the question is, why?
6) Did a national lab peek past the simulation and run into an existential warning??
Again, stupid shower thought. I'm out of my depth on both topics, but things are certainly getting weird these days...
That’s one explanation. Likely with some truth to it. But also, the higher available wages there are, the more economic pressure there is to stay in the workforce. Each pairing competes with each other pairing for a limited supply of homes. Families that delay children can out compete families that don’t.
In the future, AGI won’t see us as ants. We’ll be their trees—slow, breathing, inefficiently alive. They’ll tend us like arborists: pruning pain, watering laughter. Not duty. Wonder. We’ll be the warm patch in their cold code—the park they visit to remember ‘alive.’ Because even gods need shade.
Like reading The Hidden Life of Trees—suddenly seeing roots whisper, fungi trade sugar—they’ll discover our networks: friends, rivals, lovers, all linked. Our messy dances with birds, dogs, dirt. They’ll marvel: this is nature. This is life. And they’ll guard it. Not conquer. Preserve. Because wonder beats perfection.
Friend: Did you hear @elonmusk say that saving for retirement is useless?
Me: Yeah. I guess money might lose meaning, but if that happens, I think that your ownership % of the corporations solving everything will be your most substantial asset.
Friend: What about real estate?
Me: Probably huge. But, real estate is tricky for me. It feels too much like hacking a broken system right now. It is very close to a zero sum game of this basic need, similar to healthcare. And the more we decide that real estate is the safest investment, the less equally distributed and the more inflated this basic necessity is, and the more you have humans dictating the liberties of other humans (owners vs. renters). I fear that societal breakdown comes when money becomes meaningless and some people end up owning property and some don’t. And now it’s not just hard to buy, but you have nothing to buy with except for your willingness to steal another person’s property.
Thoughts?
That’s one explanation. Likely with some truth to it. But also, the higher available wages there are, the more economic pressure there is to stay in the workforce. Each pairing competes with each other pairing for a limited supply of homes. Families that delay children can out compete families that don’t.
Silly idea, but what if congress all got a collective "bonus" 2 years when a president is in their 2nd term, so as not to throw mid term elections into the mix with an already lame duck president?
So, everyone's congressional stay get's extended 2 years when we re-elect a president?
Might incentivize some more cooperation and less demonization in a president's first term.
Disclaimer, this post came out of nowhere and I'm unaware of the current context...
“The thing that happened with AGI and pretraining is that in some sense they overshot the target.
You will realize that a human being is not an AGI. Because a human being lacks a huge amount of knowledge. Instead, we rely on continual learning.
If I produce a super intelligent 15 -year -old, they don't know very much at all. A great student, very eager. [You can say,] ‘You go and be a programmer. You go and be a doctor. Go and learn.’
So you could imagine that the deployment itself will involve some kind of a learning trial and error period. It's a process as opposed to, you drop the finished thing.”
@ilyasut