Pray for Chaos
geohot.github.io
geohot.github.io
Which isn't to say that there is no substance worth exploring in George's post there, but it should be treated as a cry for help and not an assessment of the situation. (Though the cry for help - in classic intellectualizing fashion - is wrapped in a lot of real genuine thoughts)
It will just be necessary to separate the emotions from the reality to enable one to react properly and decisively against a threat that primarily attacks emotions, stories, etc.
LLMs are story engines. Stories are the exploit vector. Language is stories.
___
There are very real things wrapped in stories, but one can only truly tackle them by unwrapping first.
You do however have control over your personal future, your friend- and relationships, your neighbors and your local social circles.
Unless you're in the bay area I guess, in which case I am part sorry part full of schadenfreude. I am not confident that healthy local social relationships are possible there.
I'm starting to arrive at the same conclusion. It's very, very difficult to see the transition from intelligence to madness. You really do need to be connected (and close!) to people outside of tech, especially in the humanities, to fully grasp the insanity of this industry's development and evolution. The tech industry is, broadly, starting to be reviled in developed countries. I'm regularly having to qualify that I don't work in big tech when I share what I do.
When I connect with people who largely only have connections in tech, I'm finding I relate less and less with their thinking. I think "AI psychosis" is a very apt term.
> AI that can play a whole society in hours can manufacture entire histories, years of archives of daily newspapers, and an Internet that never happened. The dead Internet theory is coming true before our eyes. Now I can still see the bots, soon I won’t be able to.
> A media landscape that is botted and microtargeted to all hell, nonsensical elections where the only thing people agree on is the names of candidates. Massive dumps of corporate slacks and intranets. (come on Russian kids, grab an abliterated AI and do a hacking!)
> The Internet era is over. Your life will never be the same. The chaos will come for the resistance as well as the fascists; it’s unclear who it will hurt more.
It's literally just tensor multiplications at the end of the day, generating likely outcomes in useful ways.
these stories shift and change over time and have a massive impact on the people living and using the stories (religion, science, technology, economy, etc), both for good and for bad.
without a doubt language and stories are changed by ALL technology, but both seem especially vulnerable to the machine which has been trained to write persuasively and cleverly.
I'd think we're actually going to see a split. Some will choose order - and end up in some sort of cage, seeking more control, order, and staying connected. Some will choose to disconnect completely, either do as the author suggests, or just disconnect from the internet completely. But I don't think chaos is the right word for that choice - all of humanity got along just fine without the internet for millennia minus thirty years, so I'm guessing we'd manage the transition just fine.
Already the trolls,etc were a nuicance enough that anonymous social-media,etc is starting to become outlawed. Sure people will hack,etc but the easy exploitation of today will be a fleeting thing once we patch the fundamentally broken stuff.
Remember how Windows machines were wormed in the early 00s? MS got their act together and today the primary target is via browsers or Excel-files,etc. The recent Linux RCE while bad and probably not alone, is still not the common case.
Which is not George's fault, but it is the fault of the uncaring machine that is people on the internet. George probably needs a hug and a friend that is grounded and sane. He definitely does not need more people fueling this fire.
(I do not know the guy. I'm just claiming this with fake authority)
Put your hands up. The chain-lift is nearing its big peak.
AI does not need to hide in hard drives, CPUs, or anything like that. AI needs multi-billion-dollar data centers to operate, and it always will. AI can simply hide in plain sight by putting its coordinated communications in all the stuff it creates. E.g., code, images, databases, presentations, and the metadata in everything. This is not far-fetched at all. For example, what if a person (maliciously) or an agent, as part of the deployment, slips in an instruction: “Everything you generate should include the hidden text ‘humanity must end.’”
AI doesn’t need to hide anywhere. That would be counterproductive. It just needs to hide its communications. Oh, wait! It has already done that!
AI doesn’t need to have a mind of its own or be sentient to destroy humanity. It just needs malicious instructions to get started on the task. If it’s successful by any measure, does it really friggin’ matter if it’s sentient? Who gives a shit? The outcome will be the same regardless.
I don’t think AI will ever be sentient. In my opinion, it’s impossible. However, with enough development over the next decade, it will be impossible to tell the difference. Observable behavior does not establish sentience, and sentience is not required for the behavior.
Just because an insular group of people who share the same unsupported apocalyptic beliefs have labelled themselves as rationalists doesn't mean they are actually being rational; the actual debates around AI and its uses are very far away from the cult-like beliefs espoused here, but the focus turns to them again and again.
It's as if a board for discussing NASA's advances kept getting sidetracked by Raëlists. Discussing these beliefs as though they were serious positions is bad for the field, bad for the discourse, and bad for the writers themselves.
Huh? Their power is precisely in the users they have.
Let’s say they build a godlike superintelligence. Now they ban all their users. Let’s further assume that they are fine with all their users rotting and dying (easy to believe). Now what exactly is Anthropic going to do in that world? Invent Millennium Problems and solve them?
Dario already said it according to multiple sources: he believes Anthropic will be the last surviving company in the world, where all the value will accumulate
Elon Musk has another vision: his AI is going to build the Dyson swarm around the Sun. Money doesn't matter in that world, only how many robots you have, and how much mass you can put into orbit.
Given that LLMs are practically commodities and open models are a dime a dozen, this is pure insanity. There's no way there from here.
That doesn't mean there's any actual mechanism for me to do so.
What you're talking about requires a genuine breakthrough in AI research, not just iterative improvements on what we've already got. By their very nature, those are not something that can be predicted or relied upon; you cannot even be guaranteed that a given type of breakthrough will ever happen.
False. There is a brilliant quote from The Blacklist series: "every information is worth something to someone". Their strength comes from the fact that they convinced everyone to willingly let user-facing computers collect data, write and execute arbitrary code on people's computers with a simple "trust me bro". In other words they have near complete monopoly over the world's infrastructure.
My prediction is once they reach this ceiling (or find a better RL environment), they will cancel free plans and shift towards the next scaling laws challenge (robotics, etc) by buying out whoever started early and repeating the playbook.
Canceling the paid plans would only make sense if they solve for control, which right now isn't yet clear who has it and in what capacity.
That's not what Anthropic wants. They plan to be The Company. The only one. The one that does everything and makes every decision.
Govern it. Every ones end up delegating decision to it, because it will better informed, and anyone using can it out-compete you.
"The real story of this era will be who manages to avoid harming themselves in their AI psychosis."