correct.
This is why we need Musk to succeed. He raised $6 billion yesterday. If we have just one ‘truthful’ AI, then that will become the dominant platform for all but silly women, children – and frankly, people from less developed cultures. š
Source date (UTC): 2024-05-29 15:36:56 UTC
Original post: https://twitter.com/i/web/status/1795841765234344283
Reply addressees: @SeanStreb
Replying to: https://twitter.com/i/web/status/1795808928049074642
RT @curtdoolittle: @Gundissemenator @darkmythos_ WHAT DOES IT TAKE TO CREATE A MALICIOUS AI?
Ok, so you gave me an idea of how I can frameā¦
Source date (UTC): 2024-05-24 12:10:36 UTC
Original post: https://twitter.com/i/web/status/1793977899865047363
I can see a future where my work is on a chip that watches the memory of ai’s and can either alter weights, pull memory, or pull tasks, or pull power – just like you and I can stop ourselves from thinking unpleasant things, or just as we can stop others from doing unpleasantā¦
Source date (UTC): 2024-05-23 20:33:28 UTC
Original post: https://twitter.com/i/web/status/1793742062052941839
Reply addressees: @Gundissemenator @darkmythos_
Replying to: https://twitter.com/i/web/status/1793739111926518062
Yes well, look up the ‘paperclip’ problem and it’s a decade old and a trope in the ai field. But even if just read through it, the AI needs to be told to do it, has to have the resources to compute it, has to continue to use the resources to compute it continuously, has to
For example I have a ‘friend’ who worked in programming nuclear reactors. The thing is that there are all sorts of safeguards limiting what the computers can do to nuclear reactors and the constant capacity for human intervention even if those safeguards fail.
No one is going to ‘trust’ ai’s to do much until that risk profile is even lower than it is for ordinary human generated code today.
–“
What is the paperclip apocalypse?
The notion arises from a thought experiment by Nick Bostrom (2014), a philosopher at the University of Oxford. Bostrom was examining the ‘control problem’: how can humans control a super-intelligent AI even when the AI is orders of magnitude smarter. Bostrom’s thought experiment goes like this: suppose that someone programs and switches on an AI that has the goal of producing paperclips. The AI is given the ability to learn, so that it can invent ways to achieve its goal better. As the AI is super-intelligent, if there is a way of turning something into paperclips, it will find it. It will want to secure resources for that purpose. The AI is single-minded and more ingenious than any person, so it will appropriate resources from all other activities. Soon, the world will be inundated with paperclips.
“–
Reply addressees: @Gundissemenator @darkmythos_
Source date (UTC): 2024-05-23 20:31:37 UTC
Original post: https://twitter.com/i/web/status/1793741598632607744
Replying to: https://twitter.com/i/web/status/1793737419042546024
Yes. That’s the issue. But we would have to have the ability to produce such a thing (we don’t, and it would require vastly more sophisticated hardware that is only now in research at any scale), and we would have to make all the mistakes I’d mentioned. Not that someone wontā¦
Source date (UTC): 2024-05-23 20:26:57 UTC
Original post: https://twitter.com/i/web/status/1793740420985336053
Replying to: https://twitter.com/i/web/status/1793736535449514256
I don’t know much that’s useful. I received a notice from notify@x.com and further notices from support@x.com. That’s all I’ve got.
The emails said:
“If you think weāve made a mistake, you can appeal the violation by going to your Tweet and selecting āLearn moreā.
I don’t recall what happend when I did that although I did appeal it and they accepted the appeal.
I’ve only had one tweet ‘limited in reach’ and while it’s just the science of race differences in crime, it does technically violate their rules as they define them.
Reply addressees: @hist_legends @ALittleCrazy821 @BrismaLatr54501 @_romeko
Source date (UTC): 2024-05-23 15:35:09 UTC
Original post: https://twitter.com/i/web/status/1793666990734774272
Replying to: https://twitter.com/i/web/status/1793659642892992531
WHAT DOES IT TAKE TO CREATE A MALICIOUS AI?
Ok, so you gave me an idea of how I can frame this issue. So thanks.
So given that AI’s can take on the frame of reference and personality of whomever you want (ChatGPT can take on mine from the content I have published on the web), that means there is no underlying personality bias or framework other than the one you ask it to adopt. So if it has access to serial killer writings it can adopt the personality (values) of a serial killer. This is true.
But
(a) Since my work (our organization’s work) consists of providing universal manners, ethics, and morals that can be trained into it and NOT ignored. Even if we modify the demonstrated interests per culture, particularizing those ethics and morals for that country.
(b) Since no matter what these ai’s can spew out, that doesn’t mean that actions can be spewn out.
(c) Since there is no global state only user and task states, and since there is no global state possible, nor reason to give it a global state across users (or the compute available or even possible to do it), there is no ‘mind’ in the AI that has any ambitions other than those provided by the asks from users, or those that are given to the AI. In other words it has no ‘mind’ or ‘motivation’. It has only data.
(d) So the problem isn’t the, AI’s it’s the problem of people with lots of money, lots of compute, actively creating a constantly-running, iterative, state-memory voluminous, process (mind), with specific motivations, attaching AIs to machinery that can perform actions in the real world.
(e) So ai in general isn’t a risk. The risk is people, and it always will be people, because people always have motivations, and people always have minds, and people can always act.
Thanks for the ‘pressure’ to answer. š
CD
Reply addressees: @Gundissemenator @darkmythos_
Source date (UTC): 2024-05-23 15:26:07 UTC
Original post: https://twitter.com/i/web/status/1793664714624401408
Replying to: https://twitter.com/i/web/status/1793509107850469627
If you think I’m wrong give me an example. I have the same issue with the safety people. All I see is people trying to get compute and resources and money to do research on what we already understand – unless you give an AI permission to perform real world actions through someā¦
Source date (UTC): 2024-05-23 14:44:14 UTC
Original post: https://twitter.com/i/web/status/1793654176431378917
Replying to: https://twitter.com/i/web/status/1793508496639762522
I watched the video. There is no substance in it that articulates why an AI, especially given the architecture we are using, could do anything physical that we cannot physically interrupt. Plus it’s terribly easy to produce an AI that monitors the externalization into actionā¦
Source date (UTC): 2024-05-23 14:42:57 UTC
Original post: https://twitter.com/i/web/status/1793653853222527339
Reply addressees: @Gundissemenator @darkmythos_
Replying to: https://twitter.com/i/web/status/1793508496639762522
That would require we grant the ai the capacity to rewrite code.
Source date (UTC): 2024-05-23 13:22:51 UTC
Original post: https://twitter.com/i/web/status/1793633693053104631
Reply addressees: @Gundissemenator @darkmythos_
Replying to: https://twitter.com/i/web/status/1793508496639762522